Cooperation Stimulation Mechanisms for Wireless Multihop Networks: A

Survey
Normalia Samiana,, Zuriati Ahmad Zukarnaina , Winston K. G. Seahb , Azizol Abdullaha , Zurina Mohd Hanapia
a Faculty

of Computer Science and Information Technology, Universiti Putra Malaysia, 43400, UPM Serdang, Selangor, Malaysia
of Engineering and Computer Science, Victoria University of Wellington, Wellington 6012, New Zealand

b School

Abstract
In wireless multihop networks such as wireless sensor networks (WSNs) and mobile ad hoc networks (MANETs),
nodes have to rely on their peer neighbours in transmitting packets to intended destinations. A successful rate of
communication in these networks is assured if all nodes in the network fully cooperate to relay packets for each
other. However, due to the existence of nodes with various motives, cooperativeness cannot be ensured and the
communication goal is not achieved. Consequently, many cooperation stimulation approaches have been proposed to
address node selfishness by using, broadly, incentive-based and punishment-based approaches. These schemes consist
of several components including monitoring mechanisms, that need to be optimized in order to provide effective
ways to detect and manage selfish nodes in the networks. This article summarizes existing cooperation stimulation
mechanisms and discusses important issues in this field such as false judgment and node collusion, whereby the root
of these kinds of problems originates from the inability to obtain accurate evaluation on the behaviour of a node.
Keywords: Cooperation, Selfish behaviour, Wireless multihop networks, Incentive schemes, Punishment schemes,
Game theory, Accurate behavioural evaluation, False judgment, Node collusion

1. Introduction
Wireless multihop networks consist of autonomous nodes that perform network tasks in a self-organizing manner.
The networks, which is categorized as infrastructure-less includes several class of networks such as wireless sensor
networks (WSNs), mobile ad hoc networks (MANETs), wireless mesh networks (WMNs) and vehicular ad hoc networks (VANETs) [1, 2]. As they are infrastructure-less, there is no central controller or wired backbone to handle the
communication process. Therefore, the nodes in these networks have to act as routers and hosts simultaneously and
utilize hop-by-hop packet transmission. Consequently, with these processes occuring in the network, high frequency
of communication among peer nodes is expected. Examples of applications utilizing wireless multihop networks for
data transmission exist in military battlefield, catastrophe recovery, event monitoring sites, vehicular ad hoc technology system, health monitoring services and many other civilian applications. As applications based on wireless
multihop networks have evolved rapidly, much attention has been given to address issues that may affect the performance of such networks. These include issues related to resource constraint, security, cooperation and connectivity
where the main concern is to ensure that data can be efficiently transmitted across the networks to reach intended
recipient. In the case where packets that need to be transmitted are between nodes within direct wireless transmission
range, a successful rate of data transmission is assured if connectivity is good. However, when the source and destination nodes are beyond direct wireless transmission range of each other, the source node has to rely on relay nodes
or intermediaries to help forward the packets towards intended destination node.
author. Tel: +60389471777
Email addresses: normalia@upm.edu.my (Normalia Samian), zuriati@upm.edu.my (Zuriati Ahmad Zukarnain),
winston.seah@ecs.vuw.ac.nz (Winston K. G. Seah), azizol@upm.edu.my (Azizol Abdullah), zurinamh@upm.edu.my (Zurina Mohd
Hanapi)
Corresponding

Preprint submitted to Journal of Network and Computer Applications

March 12, 2015

When dealing with relay nodes in wireless multihop networks, the issue of uncooperative forwarding behaviour
will most likely occur because each node has different motivations and intentions. The intentions are very subjective
and may result in good, selfish or malicious behaviours. Ideally, if each node shows a good behaviour by cooperatively
relaying packets from other nodes, an optimum network performance can be achieved. However, such assumption is
not valid in the real environment of wireless multihop networks where nodes are autonomous and most importantly,
have resource constraints. With these characteristics, nodes are compelled to preserve their own energy to sustain in
the network, which is why cooperative relaying is difficult to achieve. This behaviour is known as selfishness and
can exist in any node, affecting topology design [3] and network operations like routing and packet forwarding [4].
The level of selfishness can be very abstract, making it difficult to measure and quantify because unlike a node with
malicious intent, selfishness may be shown intermittently depending on the network scenarios that it needs to handle.
For example, if a node is handling heavy forwarding loads due to its location being the closest to a particular hectic
destination, it may decide to favour certain requesting nodes or drop any incoming packets that arrive after certain
forwarding periods to avoid excessive energy depletion. From the energy conservation principle, such behaviour is
reasonable for a nodes own benefits but if it is done excessively and unfairly, the final effect towards the network may
be the same as those by malicious nodes.
In contrast to malicious nodes which intentionally plan to disrupt network services through harmful attacks, selfish
nodes do not mean to harm the network. However, the effect of selfishness may eventually cause network disruption
if all nodes decide to be self-centred by not forwarding packets for other nodes. Although selfishness can be classified
as a type of malicious behaviour, it may not be effectively overcome with security approaches. Generally, security
approaches are often implemented using cryptographic-based mechanism in order to thwart bad nodes from the network such as in [5, 6]. Although this kind of security mechanism may work well in reducing selfishness, it has high
computational overheads and is not a suitable countermeasure for selfish nodes, which are not as malign as malicious
nodes. Hence, cooperation mechanisms have been introduced to solve the problem of selfishness by detecting nodes
that show uncooperative behaviour and provide a means to discourage selfishness (i.e. stimulate cooperativeness)
using either incentive-based or punishment-based mechanisms.
The main contribution of this article is to present a comprehensive study on the existing cooperation stimulation
mechanisms by investigating the strengths and weaknesses of their key structures and elements. Using comparative
analysis, we have identified crucial issues that revolve around the selfishness detection aspect of this specialized field,
which is rooted in the inability to accurately evaluate node behaviour. The remainder of this paper is organized as
follows. Section 2 presents an overview of research on wireless multihop networks with a focus on cooperation.
Discussion on cooperation stimulation mechanisms is presented in Section 3 where existing schemes (i.e. creditpayment, reputation, hybrid and punishment schemes) are reviewed and discussed from several important angles.
This section also discusses game theory, which is the study of mathematical models of conflict and cooperation
between intelligent rational decision-makers [7] and widely applied in the study of cooperation in wireless networks.
In Section 4, a comparative analysis based on overall structure and elements of cooperation schemes is carried out and
summarized in Table 3. We extracted several important features from each existing scheme, identify strengths and
weaknesses for possible enhancements. We also provide discussion on game theory of perfect/imperfect public/private
monitoring as an alternative to available observation tools. Section 5 discusses the open research issues of this field
specifically on false judgment and node collusion problems, and highlights the main reasons behind the prevailing
problems. This section also provides some insights on what should be anticipated for future directions of node
cooperation mechanisms for wireless multihop networks. Section 6 concludes this paper.
2. State of the Art of Wireless Multihop Networks
Extensive research have been conducted to overcome the issue of cooperation in wireless multihop networks. As
depicted in Figure 1, this issue can be investigated based on two broad approaches: (1) full cooperation and (2) partial
or zero cooperation, as shown in Figure 1. These two categories are described further in the following subsections.
2.1. Full Cooperation
In this category, all nodes are considered to be fully cooperative in any communication operation. This has been
the fundamental assumption in wireless multihop networks where research focused on the improvement of relaying
2

Wireless Multihop
Networks

Resource
Constraint

Security

Cooperation

Partial/Zero
Cooperation

Incentive-based
(Payment,
Reputation)

Connectivity
and Capacity

Coverage

Full
Cooperation

Punishmentbased

Relaying
Schemes

Routing
Protocol

Figure 1: State of The Art of Wireless Multihop Networks.

schemes and routing protocols, as depicted in Figure 1. Research on the improvement of relaying scheme proposes
cooperative relaying techniques, also known as cooperative communication (CC), and the introduction of extensions to
improve performance. In the work by Herhold et al. [8], the performance of conventional relaying versus cooperative
relaying in multihop wireless networks was analyzed and compared. Conventional relaying occurs when a node only
listens or obtains information from preceding relay node and thus, the destination node only gains information from
the last relay in the transmission path. Meanwhile, cooperative relaying contributes information to the destination
by putting together all information from each relay node. The performance is measured based on how these two
techniques (as also compared to direct transmission) impact communication reliability in term of path loss, different
relay positions and spectral efficiency. Sharma et al. [9] proposed an improvement for CC in wireless multihop
networks by looking at the issues of joint routing and relay node selection during multiple transmission sessions at
one particular time. In a multihop session, when two or more traffics are initiated at the same time, tasks overlap will
most likely occur. For instance, a source node that initiates a communication session can simultaneously become a
relay node when another transmission session requires it to be an intermediary, causing a cross-function problem, thus
reducing throughput. The authors utilized a mathematical model of branch-and-cut framework and proposed few new
components (i.e. Feasible Solution Construction (FSC) algorithm, cutting plane and branching variable) within the
framework in order to optimize the effectiveness of CC. The goal of implementing the framework was to minimize
task redundancies to achieve optimum throughput.
In [10], the issue of relay node assignment in multiple source-destination nodes under cooperative ad hoc networks
was highlighted. The specific problem occurs when more than one source-destination pairs want the same set of relay
nodes to transmit data. When this occurs, the rate of data transfer is low due to non-optimized utilization of available
relay nodes which become wasted resource. In order to achieve efficient relay node allocation, the authors proposed a
polynomial-time algorithm named Optimal Relay Assignment (ORA). The algorithm generally works by maintaining
linear complexity of the relay assignment in each loop, whereby available relay nodes are distributed evenly to the
source-destination pair nodes, thus increasing data transfer rate for each pairs as proved by the simulation results of
this paper.
On the other hand, the full cooperation approach has also been investigated from the perspective of routing protocol
enhancement, which aims to increase the efficiency of communication operation. In [11], the authors aim to provide
an efficient communication operation in multi-source multi-destination multihop wireless networks by proposing a
distributed routing scheme with collision avoidance strategy. Their paper focused on collisions that occur when two
or more source nodes transmit data over one intermediate node at the same time. The authors suggested that instead
of allowing collisions at the intermediate node, there should be an alternative node to carry out the transmission task.
3

Thus, they introduced the concept of virtual node and virtual link. A virtual node consists of a collection of several
nodes that cooperatively receive or send data to a receiver node, whereas a virtual link is the link that is associated
with the virtual node. By having a virtual node, nodes within the group would cooperatively decide on which node is
more deserving of the transmission process. Hence, collision and bottleneck could be avoided.
In [12], the authors proposed an extensive framework of cooperative protocols for wireless ad hoc networks. They
focused on utilizing cooperative relaying in three layers of OSI: physical, data link and network. For physical layer,
unlike most works that only considered one type of scheme, this work provided the flexibility of using repetition
or space-time coding scheme. Furthermore, the coding scheme is supported by proposed MAC extensions, which
are maximal ratio combining-MAC (MRC-MAC) for repetition coding and space-time coding-MAC (STC-MAC) for
space-time coding in data link layer to support cooperative relaying. On the other hand, a cooperative routing protocol
(CRP) has been proposed for network layer to ensure reliable transmission from source to destination node. Since
AODV routing protocol was used as a platform, the authors also proposed an enhanced-CRP (E-CRP) to solve the gray
zone problem in AODV, which occurs when data packets are not transmitted after a valid route has been established.
In another study [13], the goal was to produce an energy-saving communication process in wireless network (at
physical and network layers) through cooperative routing mechanism. The authors proposed an optimal cooperative
routing algorithm through dynamic programming (DP) concept and introduced two sub-optimal algorithms to minimize energy-usage in route establishment process. The algorithms limited the number of transmitted messages to one
in a particular transmission session that help to attain optimal energy-saving by selecting the minimum-used-energy
route. However, the study did not consider the effect of having more than one message being transmitted in a route
and the existing of multiple flows in the network.
2.2. Partial/Zero Cooperation
Partial/zero cooperation is a condition where fully cooperative behaviour cannot be guaranteed. The scenario is
especially likely for ad hoc networks where mobile devices have portable power sources with limited amounts of
energy, which drives nodes to behave selfishly. Under partial or zero cooperation, the approaches that have been
proposed are classified into two major categories, namely, incentive-based and punishment-based mechanisms. Both
approaches have been widely implemented in wireless multihop networks, specifically to handle misbehaved node in
the network layer by obtaining information on node forwarding behaviour prior to taking necessary actions.
Two main types of misbehaviours exhibited by nodes are maliciousness and selfishness. Malicious is an attribute
of nodes that intentionally disrupt network operation by launching either passive or active attacks, severely paralyzing
the network. However, such an action is usually visible and can be controlled by using security countermeasures. It is
the latter (i.e. selfishness) that is trickier to capture because it is rarely displayed explicitly by a node. A selfish node
does not have the intention to flood the network with malicious injections, but the act of abstaining to cooperate may
also collapse the network. In other words, network disruption is a targeted goal of malicious nodes but it is a final
outcome of selfish nodes. In order to handle selfishness, cooperation stimulation mechanisms have been introduced
in many existing related works and detailed discussion is presented in Section 3.
3. Cooperation Stimulation Mechanisms in Wireless Multihop Networks
The direction of discussion for this section is based on the two major classifications of cooperation stimulation/selfishness mitigation mechanisms, viz., incentive-based and punishment-based mechanisms/schemes as shown
in Figure 2. The main goal of these proposed schemes is to enforce cooperation of nodes to achieve optimal network performance that can be measured using performance metrics like throughput, packet delivery ratio, accurate
behaviour detection rate, etc.
3.1. Incentive-based Mechanism
Incentive-based mechanism is an approach that has been proposed to provide motivations for nodes to cooperate
in the form of credit payment or reputation award (viz. good or bad), following the observation and evaluation on the
forwarding behaviour of a particular node.

Cooperation Stimulation /
Selfishness Mitigation

Incentive-based Mechanism
(1) Credit payment scheme [14-26]
(2) Reputation scheme [27-42][46]
(3) Hybrid scheme [45][47-48]

Punishment-based
Mechanism
[30-32][49-53]

Misbehaviour Observation
and
Detection

Passive
Acknowledgment

Promiscuous
Mode
[27][29-33]
[39][40-42][46]

Active
Acknowledgment

Peer Nodes
ACK

Probing
Technique

[34-38][45]

[70-71]

Specificationbased Approach

Game
Theoretical
Approach

Perfect
Imperfect
Finite State
Public/Private
Public/Private
Machine Model Monitoring
Monitoring
[72-76]

[49][51-52][59]
[62-63]

[50][82]

Figure 2: Classification of Cooperation Stimulation Components

3.1.1. Credit Payment Scheme

The concept of virtual credit has been introduced under the payment scheme, which a node holds as a way to
represent their cooperativeness. The amount of credit will increase when a node helps to forward other nodes packets
and decrease when it refuses to cooperate. When a node runs out of credit, it will not be able to send its own packets.
Thus, collecting credit by forwarding packets for others becomes the motivation for nodes to be cooperative. Method
of payment for nodes can be viewed from two main aspects, viz., distributed and centralized architectures, whereby
reliance on tamper-proof hardware and central-agent are involved respectively.
Distributed Architecture
In terms of the scalability aspect, a distributive payment method offers advantages for nodes to administer their own
credits such that prompt transactions can be done more conveniently using the tamper-proof device installed in each
node; credits are charged and paid according to nodes sending and forwarding operations without having to rely on a
third-party manager. Examples of credit schemes that implement tamper-proof device are [14, 15].
A virtual currency called nuglets rewards nodes that cooperatively help in forwarding packets [14]. The payment
system worked based on two models known as Packet Purse Model (PPM) and Packet Trade Model (PTM). In PPM,
a source node pays all its forwarders along the path to destination by appending sufficient nuglets in the packet. Each
intermediate node will take some nuglets according to its forwarding cost and if there is no nuglet left, the packet
is dropped. Hence, it is important for a source node to avoid under estimation of nuglets. In PTM, nuglets are not
appended in a packet, instead they are traded via buying and selling activities among intermediaries in order to forward
the packet where the destination node will eventually pay for the total cost. This method helps to reduce the number
of dropped packets due to under estimation of nuglets by the source node, but it may cause node to lose their nuglets
unnecessarily, such as, when intermediaries refuse to buy the packets (i.e. being uncooperative). The scheme was
enhanced in [15] by introducing credit-counter to reduce overloading of packets by a source node. The counter will
increase when a node helps in forwarding packet for others and decrease when it sends out its own packets. Thus, a
node must balance its forwarding activities, so that the counter will remain sufficient in order to survive in the network.
5

Despite having the ability to provide flexibility for nodes to manage credits, distributed-based credit scheme does
not gain wide adoption due to the following factors:
Cost - as the name implies, the tamper-proof device is claimed to be resistant from being tampered by any
forgery attacks which reflects costly hardware. It might not be practical for every node to have such a security
device considering that the usage might not achieve optimum level.
Resistance Level - although the device might be durable from external attack, it can still be manipulated by its
owner to collude with other nodes and perform fraudulent transactions. In certain cases, it has to be assumed
that the device could provide optimal security protection [16] but such an assumption might not hold in many
real scenarios.
Unfairness - due to nodes dispersed locations, getting fair opportunity to obtain credits for continuous network
operations is hard. Nodes which are located in the middle of the network will have more opportunities to forward
packets for other nodes (gain credits), whereas, nodes on the periphery of a network have smaller chances to be
selected to participate in the communication process (lose credits). When credits cannot be distributed evenly in
the network, resources (i.e. potential cooperative forwarders) have the tendency to be wasted, thus, cooperation
cannot be ensured.
Excessive Overheads - a node needs to forward enough packets to gain credits for its own packets to be sent
out and possibly forwarded by other nodes. In the event of major packet loss or dropped packets, the amount of
credits earned by a node may not be sufficient to cover the cost of re-sending its packets. In this case, a node is
charged more, with no guarantee that its packets will be forwarded optimally.
Centralized Architecture
Another method of credit payment is based on centralized architecture where reliance on central agent (i.e. bank)
to manage credit transactions is involved. Credit is paid before or after a particular node is charged or rewarded,
by submitting receipts/reports to the central agent for verification and issuance of payment. Receipts submission
can either be done online (on-the-go transaction with the bank) or offline (accumulated transactions sent in batch to
the bank). Having reliance on a central agent removes the need for each node to manage credit transactions, thus,
reducing the processing burden and the need to have tamper-proof device. Nonetheless, this approach comes with
several drawbacks whereby bottleneck problem during communication with the central agent is the most pronounced
issue. We will first briefly discuss existing schemes based on this approach prior to presenting its pros and cons.
In Sprite [17], nodes keep a receipt of its forwarding activities and claim for payment from a central agent named
Credit Clearance Service (CCS). To prevent the payment from being pushed to the destination node by the source
or/and colluding nodes, only the source node is charged for packet transmissions it has initiated, and forwarders gain
credits only if the CCS received satisfactory proof from corresponding next-hop nodes stating that forwarded packets
have been received successfully. Anderegg and Eidenbenz [18] proposed a payment-based scheme that functions
as underlying concept in their design of routing protocol named Ad hoc-VCG. VCG, which stands for Vickrey [V],
Clarke [C] and Grove [G] adapted a pricing-based mechanism called the second-best sealed bid auction. Ad hocVCG protocol was designed with the aim of establishing a routing path that is cost-efficient and reliably truthful
where every node is required to genuinely publish its cost of energy during the route discovery session. A source
node will choose the most cost-efficient (energy-wise) and truthful (policy-obedient) route to send its data packet and
pay the intermediate nodes via the central bank. This approach is similar to PPM model proposed in [14] except
that forwarding cost of each intermediary needs to be published beforehand to eliminate the problem of over/under
estimation of cost by source node. A game theoretical technique was also proposed to ensure truthfulness. Ad
hoc-VCG was shown to be effective in finding energy-efficient and truthful route but with assumption that all nodes
genuinely declare their real cost and there is no collusion among nodes to illegally maximize payment; this was not
addressed in the work.
An enhanced version of Ad hoc-VCG, COMMIT protocol [19], allows a source node to negotiate an affordable
cost with the intermediaries rather than leave the decision to the intermediaries. In the initial route discovery process,
a source node or sender will announce the maximum total cost that it can offer if a connection is established. Upon
6

receiving packets containing such information, intermediaries will decide whether to receive or reject the offer. Assuming that all related intermediaries agree with the offered payment and that a routing path is found, the destination
will return all gathered information to sender. If the requested total payment is lesser than the offered amount, sender
would choose the route. Even though the sender gains control over the amount of payment, the processes that it needs
to go through are quite tedious and can lead to high communication overhead. In reality, it is difficult to achieve
total agreement by intermediaries on the offered amount, unless the sender can provide a flexible amount of payment.
Janzadeh et al. proposed a scheme known as Express [20], which is a credit-based mechanism that inherits some
common features of Sprite [17], except that it offers a lighter computational overhead, while providing a secure credit
transaction by adopting hash chains instead of digital signatures for authentication. The efficiency of the proposed
scheme was ensured by overcoming any cheats with necessary incentives and penalties. In [21], a secure credit payment scheme named Report-based pAyment sChemE (RACE) was proposed. RACE is a secure scheme that aims
to have low communication and processing overhead due to the dependability on central agent. It was done by the
introduction of lightweight payment reports that need to be submitted to the agent, instead of using common receipt.
The term lightweight is defined as having less information appended on the reports, yet still sufficient to provide necessary information. Extra information will be submitted to the central agent only when the need arises like when a
node requires more evidence on a cheating node. Furthermore, the evidence will only be stored temporarily to reduce
storage overhead. The efficiency of the proposed scheme was validated by their analytical and simulation results.
The centralized architecture has been receiving wider acceptance in comparison to the decentralized approach
and has been adopted in many other credit-based schemes [22, 23, 24, 25]. Although the centralized approach offers
several advantages, especially by removing the reliance on tamper-proof device, it comes with several drawbacks
(including the aforementioned unfair share problem suffered by the distributed architecture) like high communication
and processing overheads, verification delay and false verifications, increased storage overheads, and the need to
secure the central agent as well as information held by it. As both distributed and centralized architecture have
pros and cons, a hybrid scheme named CASHnet [26] has been proposed by combining the good features of both
architectures. In CASHnet, the use of tamper-proof device (via smart card) is preserved to provide flexibility for
nodes to manage their credit transactions while at the same time depending on a centralized service station to top-up a
nodes credits. Although CASHnet is able to maintain credits in the network in the long-run, it still suffers from high
protocol overhead.
3.1.2. Reputation Scheme
Reputation is another type of incentive used to stimulate cooperation in wireless multihop networks. A node collects the past behavioural information of other nodes based on its own observations, reports from trusted neighbours,
or both. The information, indicating their level of selfishness, is used to determine the reputation value of monitored
nodes. A reputation level lower than a predetermined threshold value may reflect that a particular node is selfish,
thus will be avoided during the communication process. The most essential element in measuring reputation is to
determine a node behaviour via techniques like passive and active acknowledgment (cf: Section 4.1). Once a node
behaviour has been determined, it will be passed to the reputation system components for further evaluation.
Rating/Voting Mechanism
Rating/voting is an essential element in a reputation scheme whereby an observer node rates the level of cooperativeness/selfishness of a node in routing and forwarding packets. Subsequently, the rating may be shared in a voting
system to further evaluate a nodes behaviour by collecting opinions from other nodes prior to enforcing punishment
towards a selfish node. Among important aspects that need to be considered under this element are how to rate a
nodes behaviour, how to process votes from nodes to compute the reputation level, how to share the votes effectively
and how to ensure that the shared votes are not falsified.
Early work adopting the rating mechanism for reputation involved monitoring nodes activities using a watchdog
mechanism [27], in which the information gathered is used to rate a node such that a node with low rating is excluded
from the routing path (pathrater). The work in [27] is considered as the pioneer in utilizing promiscuous mode of operation to obtain a nodes forwarding activities, whereby a watchdog node observes (i.e. listens) the next hop neighbour
in order to know whether it helps to relay received packets or simply discards them. Continuously discarding packets
causes the misbehaving rate of a particular node to increase and when the rate surpasses a predefined threshold value,
7

it is labeled as misbehaved and is avoided in the next route establishment process as a form of punishment. However,
such an action will actually benefit the misbehaved node in the sense that its forwarding burden is lessened by not
committing to the communication process initiated by other nodes.
Reputation schemes that had adopted promiscuous mode as a monitoring tool and used rating/voting include
CONFIDANT [28], CORE [29] and SORI [30]. CONFIDANT (Cooperation of Nodes: Fairness in Dynamic Ad
Hoc NeTworks) [28] aims to develop a better path rating mechanism. Misbehaviour is detected by neighbourhood
nodes where information is sent to a reputation system (i.e. voting mechanism) that collects sufficient evidences and
votes from observing nodes. Collaborative Reputation (CORE) [29] was introduced to overcome the problem of node
falsification in CONFIDANT by restricting dissemination of negative rating across the network, which means that
the mechanism only allows nodes with good rating to be announced so that any attempt of falsification by colluding
nodes remains in vain. However, false announcement could still happen when nodes collaboratively rate a bad node
as good, and thus, impairing the credibility of the announcement. In both of these schemes [28, 29], the reputation
information is broadcast globally causing high communication overheads. Secure and Objective Reputation-based
Incentive (SORI) [30] was introduced to reduce the overhead by sharing information on the reputation of a node with
its neighbouring nodes only.
On the other hand, work on mitigating the adverse effect of globally shared reputation information have been
proposed by emphasizing and enhancing the ability of local observation. In [31], a scheme named Observationbased Cooperation Enforcement in Ad hoc Networks (OCEAN) was introduced to fully utilize the advantage of firsthand/local observation to avoid global exchange of reputation information that is prone to manipulation by colluding
nodes. Each node keeps the rating information of its neighbours in the same wireless channel area, where the default
rating value is neutral. The value will either increase or decrease according to the forwarding actions taken by the
neighbours. If the rating value of any neighbour falls below a certain threshold, that particular node will be added to
a faulty list and, as punishment, any request for services from this node will be rejected; a decision made locally by
an observing node. A unique feature of this scheme is the introduction of a second chance element whereby after a
punished node has been inactive for a certain amount of time, it will be moved from the faulty list and given another
chance to increase its rating by being cooperative. This element was found to help in reducing false judgment, where
any node that was mistakenly labelled as misbehaved obtains another chance to vindicate itself.
OCEAN was proven to perform better than a generic scheme which utilizes globally shared reputation information
due to the reduction of node collusion risk. However, the scheme is still unable to cope well with the reliability of a
local decision made by any node, as it is possible that a node falsely accuses other nodes as misbehaved. In addition,
the faulty list can also be tampered by attacker. The weaknesses in OCEAN were later addressed by a scheme
called Hybrid Mechanism to Enforce Node Cooperation (HEAD) [32] which utilizes warning messages instead of
relying on broadcasting faulty list during the route discovery phase. The scheme proposed in [33], known as Locally
Aware Reputation System (LARS) also emphasized on the use of local/direct observation rather than globally-shared
information on reputation, which can cause high communication overhead and the risk of colluding nodes. Once
an observed nodes reputation value falls below a predetermined threshold due to the act of dropping packets, its
selfishness will be identified, and a warning message (signed with signature) of the behaviour will be signaled to
k-hop neighbours.
Another approach involves the use of end-to-end and hop-by-hop acknowledgment mechanism to overcome the
issue of colluding nodes giving false accusation or false praise on a node behaviour [34, 35, 36]. On the other hand,
hop-by-hop acknowledgment such as [37, 38] was introduced to reduce the adverse effect of accusing whole link as
unreliable imposed by end-to-end acknowledgment method.
Trust Management
In determining the threshold level of bad and good reputation, node behaviour may be assessed by its trustworthiness
level which can be derived based on trust features or evidences. There have been several methods proposed to compute trust metrics to evaluate node trustworthiness and thus, determine node selfishness that will be discussed in the
following existing existing schemes.
To obtain stronger evaluation of a node behaviour, CONFIDANT [28] introduced a trust manager that crosschecking a particular misbehaviour report prior to making announcements. However, this scheme is susceptible to
false accusation by colluding nodes towards a well-behaved node. Besides, the reliability of the trust managers
opinion is questionable due to its vague functionality. Chong et al. proposed a detection framework named Separation
8

of Detection Authority (SDA) [39] to improve the trustworthiness of shared information on behaviour by delegating
the task of detecting selfish nodes to three different entities, named reporters, agents and central authority. When
a node (i.e. reporter) finds out about the selfish behaviour of a particular node, it submits a report to the central
authority that is going to investigate the report by appointing agents (i.e. neighbours of the suspected nodes) that
would also observe the node. Assuming that all agents would submit their reports to the central authority, a final
evaluation is made based on majority votes by the agents. Although this approach seemed to succeed in detecting
selfishness by having a central authority to validate reports, the process is costly due to the need to send and receive
additional investigation packets. In addition, due to the need for second-hand observations, SORI and SDA share the
same weaknesses as CONFIDANT and CORE such that a reputable node may falsely report a node as having the
correct behavioural state (i.e. falsely accuse or falsely praise [40]) and having the erroneous information on reputation
exchanged among nodes.
A judiciary system inspired trust model, known as Trust-based Exclusion Access-control Mechanism (TEAM) [41],
has been proposed and is made up of two-tier modules named local and global context. The local context consist of
nodes (i.e. witnesses) that collect behavioral information of their one-hop-neighbours (i.e. defendant) based on selfobservation (using promiscuous mode of listening) and recommendations from peer-trusted-nodes. The collected
trust level information will then be passed to the global context module, consisting of jury nodes, that will further
evaluate the evidence using a voting mechanism based on majority votes. Subsequently, nodes that are found to be
malicious/uncooperative, will be punished and denied access to the network. The main goal of TEAM is to provide
an accurate exclusion mechanism of misbehaved nodes with low message overhead which was proven based on their
analytical and simulation evaluations. However, some of the proposed elements are still obscured in the sense that the
trustworthiness of jury nodes cannot be firmly assured prior to getting their opinions. Although the model practiced
random selection of jury approach to avoid node collusion, the reliability of information provided by the jury, must be
validated beforehand. A mechanism that also utilized the trust model was proposed in [42]. Significant difference of
the trust model in this work is that, the model is strengthened with trust features, viz., trust value metrics, packet precision and blacklists. The trust model was implemented in their proposed friendships mechanism such that only friends
(i.e. nodes) that fulfilled the trust features criteria will be fully trusted and used in the route establishment process.
Thus, the trust model in [41] may need to consider using trust features and friendships mechanism to strengthen the
reliability of the selected jury nodes.
In [43], the trustworthiness of a node is modeled by calculating the weighting average of a particular observed
node performing some network tasks over the course of accumulated observations. The accuracy of trust value is
strengthened by integrating the model with link quality information measured using statistical model. Although the
model was developed for securing routing protocol against several malicious attacks, but it might as well be adopted
by any reputation schemes to evaluate trust and combat node collusion. An approach to evaluate the compatibility of
existing trust metrics with trust-based routing protocols using routing algebra has been proposed in [44]. The analysis
performed in this work contributed to the discovery of several conditions under which the proposed trust-based routing
protocol works correctly and optimally from the aspects of such as diversity and absorptive. Also, this work presented
cases of where certain trust metrics usage would induce conditions like routing anomalies and inference problem that
should be avoided in any trust-based routing protocol design; including reputation-based scheme.
3.1.3. Hybrid Scheme
As an alternative to the two types of incentive mechanism, the hybrid scheme was introduced, featuring the
advantages of both credit payment and reputation schemes. The issues that have led to the creation of hybrid schemes
include even distribution of credits, challenges of having a central agent in volatile network environments like MANET,
minimizing processing burden in nodes if a decentralized architecture is adopted, and dealing with node collusion.
In what follows, these proposed hybrid schemes attempt to address the aforementioned issues. A hybrid scheme
named hybrId inCentive mechAnism for coopeRation stimUlation (ICARUS) [45] was proposed to improve unfairness
distribution of credits among distant nodes by utilizing reputation estimation to control credit exchange. ICARUS
inherits some features from a reputation scheme called DARWIN [46]. This scheme relies on a central agent, named
ICARUS Credit Accounts Service (ICAS), to manage all credit transactions by acting as reward coordinator, and
eventually determine a selfish node. However, it was not explained how efficiently can ICAS collect the needed
information. Nodes are assumed to be willingly bounded to follow cooperative policy endorsed by the ICAS, which
in reality is difficult to achieve in ad hoc networks due to the autonomous nature of the nodes. In addition, reliance
9

on a central agent and acknowledgment packets dissemination processes are computationally costly which can incur
high communication overhead. These limitations were noted but not rigorously analyzed.
ARM [47] is another hybrid scheme based on the combination of reputation and credit systems. Unlike ICARUS,
it does not rely on a central agent to manage credit transactions. Instead, each node has to organize credit payment
among corresponding peer nodes based on the value of their reputation. A significant difference between ARM and
non-hybrid schemes with the same technical approach is that the load of storing reputation information is assigned to
nodes that are static or in low mobility state and are assumed to be relieved from relaying tasks. These nodes act as
reputation managers to save mobile nodes that function as relays from being burdened by heavy computational loads.
However, the authors did not clarify how certain nodes are regarded as static or low mobility, thus are appointed as
reputation managers. Although such an approach may have reduced the heavy processing burden on nodes, freeing
some nodes from relaying tasks may not be a good way of utilizing resources.
A scheme named reputation-based credit mechanism (RCM) [48] was introduced with the aim of using the reputation level of a node to estimate the risk of payment prior to submitting it to a central agent called Settlement Centre
(SC). Although this work has been able to propose an efficient way to balance the amount of charging and rewarding
cost, it is only applicable to network environments with small number of selfish nodes.
3.2. Punishment-based Mechanism
Punishment-based approach penalizes nodes that behave selfishly by isolating them from a routing path or denying
them any communication service request. Punishment is the signal to such nodes that selfish behaviour does not benefit
them. Besides being applied when a selfish node is detected, punishment is also the next step to be taken after node
is rated with low reputation value. Thus, reputation and punishment or payment schemes always work hand in hand,
and some of the above described work on reputation mechanisms can also be classified as a punishment scheme such
as in [30, 31, 32]. For payment schemes, the punishment is more subtle in such a way that when a node is running out
of credits due to constant act of selfishness, it will lose the ability to forward its own packets.
Many existing work on punishment-based scheme applied the popular game theoretical Tit-for-Tat (TFT) strategy
to punish selfish nodes [49, 50]. In TFT strategy, a node will take similar action (i.e. cooperative or selfish) as
what the other node has done previously. The strategy is played in such a way that a node will be cooperative in
the first game stage and subsequent action depends on the opponents behaviour in the preceding stage. A packet
forwarding based game model implemented in wireless ad hoc networks was proposed by [51]. This study looked
into possible situations on whether or not cooperation in wireless ad hoc network can emerge without the aid from
incentive mechanism. In order to analyze such state, the authors utilized a game theoretic model based on Tit-for-Tat
strategy, forwarding game and repeated game. The connectivity among nodes is illustrated by strategizing network
topology and routing path based on dependency graph, while cooperation is determined based on nodes willingness
to forward packets for others. Simulation results showed that some portion of nodes can cooperate naturally, thus an
incentive mechanism to stimulate cooperation is not needed.
A rigorous study on how to stimulate cooperation in a wireless multicast network using game-theoretic approach
was done in [52]. Three main important aspects of cooperation stimulation have been proposed in this work: the
modeling of cooperative aspect using infinite repeated game concept to determine the optimal cooperativeness power
level, the introduction of punishment based mechanism based on Worst Behavior Tit-for-Tat (WBTFT) strategy and
the proposal of a realistic interval estimation mechanism when the result of node behavior monitoring is imperfect.
The key concept of cooperation stimulation strategy in this study was in monitoring results, where a node will act
equally bad toward a deviating node as a punishment for the act of selfishness (WBTFT), thus reducing the payoffs
of selfish nodes. With this mechanism implemented, nodes are theoretically expected to cooperate in order to gain
higher payoffs. This theoretical analysis was validated with simulation results and proven to achieve sub-game perfect
Nash equilibrium (NE) under certain conditions. In order to achieve realistic cooperation, the authors considered the
case where the monitoring results are imperfect due to external factors like noise and interference, and proposed enhancement of the interval estimation to achieve average payoffs that are almost equivalent to that of perfect monitoring
output.
Another variant of the punishment strategy is based on Grim Trigger which was applied in [53]. Nodes that
follow the grim trigger strategy starts the game by being cooperative and will continue to cooperate until at least one
of its opponents deviates from the cooperative mode. When this happens, nodes will stop being cooperative in the
10

subsequent stages. In the next section, we will extend our discussion on game theory and how it has been used as a
tool to analyze cooperation stimulation mechanisms.
3.3. Using Game Theory to Model and Analyze Cooperation Stimulation Mechanisms
Game theory is a mathematical tool originating from economics that can be used to analyze interactive decisionmaking processes among different rational agents (i.e. decision makers) [54, 55, 56] whereby, it predicts the exact or
most likely decisions to be made. It has become a universal tool incorporated with cooperation strategies proposed
for wireless multihop networks. A game theory model for wireless networks has three main components: (1) players
(decision makers, viz. nodes), (2) actions (choice/strategy e.g. power level setting, packet forwarding, flow control
parameter, etc.), (3) preference (utility function/payoffs i.e. high SNR, low BER, low power, throughput, delay,
etc.) [57]. In a game, each player has its own preferred strategies that are initiated in response to other players
strategies in order to gain reward. The same scenarios can be mapped to the way nodes in wireless multihop networks
communicate, in which each nodes actions are heavily dependent on whether or not it will gain payoffs as a result of
performing certain actions (i.e. cooperation). Considering that all nodes expect payoffs by being cooperative, this is
where game theory can be utilized to reduce selfishness and encourage cooperation.
Table 1 summarizes several existing game theoretical approaches (including those discussed in previous section)
addressing the issue of selfishness in wireless multihop networks, where some important key points have been highlighted. In order to solve the problem of selfishness, these schemes have implemented incentive- or punishment-based
mechanisms and incorporate them with game theory elements like repeated game, prisoner dilemma, Tit-for-tat strategy and Nash bargaining. These elements were utilized to analyze the conditions in which cooperativeness can occur,
such that nodes can be naturally encouraged to cooperate, while knowing that they will not gain payoff by being
selfish. Although game theory can be applied to solve cross layer issues [55, 58], the work presented in Table 1
concentrates mainly on cooperativeness issues in network layer like routing and packet forwarding.
For example, in [59], the work focused on resolving the issue of nodes silently drop control packets during route
request (RREQ) session in ad hoc networks. Thus, they implemented a monitoring mechanism during route discovery,
but not during data packets transmission after route has been established. The basic idea is that, the amount of RREQs
relayed should almost be equal to the number of nodes of the same neighborhood. Otherwise, it simply proves selfish
activities are on-going. If a node does not reach a pre-determined threshold value of relayed RREQ, it is regarded as
selfish and will be punished. In order to avoid false monitoring detection, the algorithm is strengthened with repeated
prisoner dilemma game. A significant contribution of this work is that the authors did not use promiscuous listening
and indirect communications to monitor node behaviour in order to save energy. Instead, they used a non-complex
fully distributed monitoring algorithm to detect and punish selfish nodes. In [60], a mechanism was proposed to model
a cooperation strategy among nodes and access point (AP) in a cooperative wireless network. They concentrated on
how to ensure each user (source and relay) can have symmetrical benefits of helping each other in the network. Hence,
a symmetrical system model based on Nash bargaining solution has been proposed. In this model, two nodes that can
both act as a source and a relay can bargain with each other on when or how much bandwidth they can provide
for cooperation to take place. In the bargaining process, if both nodes can reach a satisfactory condition, they will
cooperate. Otherwise, they will operate independently which does not benefit the network in the end.
A unique incentive-based mechanism was proposed in [61] that aims to challenge the norm of cooperation stimulation approaches by proving that full cooperation of nodes is not necessary, given certain unavoidable condition in the
network. Hence, this work proposed a power control mechanism to exploit the existence of seemingly non-cooperative
nodes but actually having unintentional factors like energy depletion that restrict them from being cooperative. The
goal is to determine an optimal point of cooperation where a certain submaximal forwarding rate is sufficient, that
allows a particular node to temporarily suspend its effort in order to restore its energy, and yet not be accused as
selfish. In this way, node is expected to be naturally willing to cooperate as it knows for sure that by being optimally
cooperative will not jeopardize its resources. The novelty of this work is undeniable, however, deeper considerations
need to be investigated as many real network scenarios with more challenging conditions have not been considered.
For instance, it is still hard to distinguish liar nodes from the genuine nodes that have stopped their efforts due to
optimal forwarding point having been achieved. Hence, further verification is required to ensure that the proposed
mechanism can indeed work efficiently.
In term of efficiency measurement, while Pareto optimal (a condition where one player/node impossibly gains
payoff without making at least one other player/node worse off) is an option, most schemes utilized Nash equilibrium
11

Table 1: Game Theoretical Approaches

Author
Felegyhazi
et al.
(2006) [51]
Huang et al.
(2007) [59]
Zhang et al.
(2008) [60]

Lee et al.
(2008) [62]

Cooperation
Mechanism
Topology
control
Incentive mechanism (certain
condition only)
Punishmentbased
Incentive mechanism (bandwidth allocation)
Incentive mechanism
(paymentbased)

Lightweight
Ng and Seah
punishment(2010) [50]
based

Focus Problem
To prove that under
certain conditions,
cooperation can exist without incentive mechanism
Detecting selfish
nodes during route
discovery

Contribution

Game Theory
Element

Efficiency
Validation

Introduce dependency graph concept

Cooperative model
Tit-for-Tat strategy
Forwarding game
Repeated game

Nash equilibrium

Selfishness detection and punishment algorithm

Repeated
dilemma

Nash equilibrium

Unfairness of benefits
Symmetrical sysdistribution among
tem model
cooperative nodes
Fairness problem in
forwarding data

Introduce
virtual currency as
reward

Unstable network
behaviour with the
existence of selfish
nodes

Propose a protocol
that is robust
against
selfish
nodes

prisoner

Cooperative
game model
Nash bargaining
solution
Complete information model
Unrepeated noncooperative game
Cooperative game
model
Infinite repeated
game (Aoyagis game)
Tit-for-tat strategy
Imperfect private
monitoring for dynamic
Bertrand oligopoly
Grim Trigger strategy

Paretooptimal

Nash equilibrium

Nash equilibrium

Zhong and
Wu (2010)
[63]

Incentive mechanism
(paymentbased)

More than one

node cooperatively
misbehave to gain
incentive by cheating

Cryptographic
scheme to prevent
collusion
and
profit transfer

Complete information model

Repeated game

Strong
Nash equilibrium

Niu et al.
(2010) [52]

Incentive mechanism (punishment-based)

Cooperation
Improvement
enforcement using
of
imperfect
imperfect monitoring monitoring

Infinite repeated
game
Folk Theorem

Absolute
fairness
Pareto
optimal

Less stringent
incentive
mechanism

To achieve transmission efficiency

with
the
existence
of
noncooperative nodes
To identify noncooperative
nodes
due to unwitting
condition like energy
depletion and node
mobility

Infinite repeated
game
Dynamic repeated
game

Subgame
Perfect
Nash equilibrium
(SPNE)

Sun et al.
(2013) [61]

Exploiting noncooperative nodes

to ensure communication
operational availability
Propose power
control mechanism
to find an optimal
point where full
cooperation is unnecessary to avoid
energy wastage

12

How efficient is the

mechanism in stimulating
cooperation?

Cooperation Stimulation
Mechanism

How to manage selfish

nodes?

Incentive-based /
Punishment-based

How to share behaviour

information effectively?

Information
Broadcast
Node Behaviour
Judgment

How to identify selfish

behaviour?
How to obtain behaviour
information?

Monitoring
Mechanism

Network throughput /
Routing overhead /
False detection

Credit / Service denial / Route

avoidance/ Redemption

Efficient flooding / Gossip

routing / Trust management
Individual assessment /
Neighborhood opinion
Passive or Active ACK /
Specification / Game theory

Figure 3: Structure and Components of Cooperation Stimulation Mechanism

(NE) to validate the efficiency of the game theoretical elements in used such as [50, 62, 63]. A perfect NE is achieved
when all nodes stick to their preferred strategies and do not change even after considering other nodes strategies.
By assuming that all nodes are rational, nodes will not deviate from steady state game strategies because their payoff
cannot be increased. Thus, in the case where cooperativeness is chosen by players as strategy, it is expected to
be maintained throughout the whole stages of a game. However, in order to achieve NE in real wireless multihop
networks, major assumptions are required including unrealistic ones. For example, it is assumed that nodes can obtain
complete information about other nodes strategies, without considering the feasibility of getting such information in
real networks. In many cases, they have to rely on promiscuous listening to implicitly obtain other nodes actions
such as in [51, 46]. In addition, most work modeled game interactions for static nodes only where the occurrence
of repetitive game is highly assured. In reality, it is challenging to ensure that a game can be played repeatedly even
for a static topology due to problems like communication disconnectivity and busy channels. Hence, the proposed
solutions may not be flexible enough to cater for the dynamic nature of nodes in wireless multihop networks. These
scenarios show that significant challenges need to be taken care of before game theory can become an efficient tool
for cooperation stimulation in real networks.
4. Structure and Components of Cooperation Stimulation Mechanisms
The overall structure and basic components that compose cooperation stimulation mechanisms/schemes are illustrated in Figure 3. Basically, there are five major components involved, whereby each component provides solutions
for arising questions listed on the left side-bar of the structure by using available approaches listed on the right side.
The two upper components have been discussed in Section 3 and summarized in Table 3 with highlights on general
strength and flaws of each cooperation-based category. The lowest layer denotes the most fundamental component
that should be present, which is the monitoring mechanism used to observe node behaviour. The behaviour information will be evaluated using node behaviour judgment approaches denoted in the upper layer, whereas, in the case
where information needs to be broadcast, approaches presented in the third layer take part. The bottom three layers
are elaborated further in the following subsections.
4.1. Monitoring Mechanism
The process of observing and detecting node behaviour plays important roles in ensuring the effectiveness of the
other components. However, getting accurate evaluation of a nodes behaviour is challenging due to several factors
that may be derived from either false judgment by observer or weaknesses of the monitoring methods used. Issues
that are being addressed include inaccurate accusation by a group of colluding nodes, insufficient behaviour evidence
13

and so forth. Some of the aforementioned problems have been addressed in current research by improving some deficiencies in obtaining behavioural evidences such as [64, 65]. Existing work on improving this element have mainly
focused on selfish behaviour detection that resides at network and medium access control (MAC) layers. At network
layer, selfishness is demonstrated by nodes in the form of route discovery refusal, reluctance (drop or delay) to forward
packets, route exploitation (using established route initiated by other node for own data transmission), excessive idle
state and unfair provisioning of services. On the other hand, selfishness at MAC layer can be observed in the form
of unfair contention channel access, backoff manipulation and false announcement of signals state. In this paper, we
focus on discussing selfishness occurring at network layer where aspects like the monitoring technique used will be
covered. In general, a selfish behaviour that occurs in a network layer can be observed using four main approaches:
(i) passive acknowledgment, (ii) active acknowledgment, (iii) specification-based approach and (iv) game theory of
perfect/imperfect public/private monitoring.
4.1.1. Passive Acknowledgment
Passive acknowledgment, also known as implicit acknowledgment, is the most commonly utilized method to observe node activities in wireless communication networks where nodes run in promiscuous mode of operation a
feature that is widely available in existing wireless communication technologies, such as, IEEE 802.11 or WiFi [66].
When this mode is turned on, a node is capable of overhearing any transmitted packets by neighbouring nodes within
its transmission range to determine their forwarding behaviour as introduced by Marti et al. [27] in a scheme called
watchdog. The watchdog mechanism has been widely applied in many work, especially those related to misbehaviour
detection and cooperation stimulation approaches. Although the mechanism could record forwarding activities by
nodes at reasonable rate, it is with assumption that channels are reliable, whereas in reality, this monitoring mechanism only works well in an ideal environment that features collisions-free communication channels, genuine misbehaviour reports and reliable transmission power. A significant advantage in using this mode is that the node does
not need to rely on information provided by other nodes in order to evaluate the behaviour of neighbouring nodes as
information can be obtained locally by listening to forwarding activities, thus reducing risk of getting false reports.
Nonetheless, it has been widely used by many existing work as shown in Table 3 for its significant success to provide
reasonable sufficient information despite the aforementioned trade-offs. This is further strengthened by other proposals to enhance the weaknesses posed by watchdog mechanism, such as [67, 68, 69], that show the viability of this
method.
Another issue that has been emphasized when utilizing this method is high energy consumption that a node has to
endure due to occasional process of interface switching. Energy consumption is an important factor to be considered
since a monitoring technique that can cause a nodes energy to deplete easily may not be a worthwhile option. Similar
concern also implies the storage overhead issue where an observing node must allocate some buffer to store the
observed information that may burden its computing resource. The combination of energy and storage usages causes
a watchdog node to be burdened with high computational tasks when performing the watchdog function. As a node
may be overloaded with other communication tasks, a technique that is as lightweight as possible should be adopted
in order to maximize a nodes lifetime, thus prolonging its ability to function in the network.
Privacy is also another issue to be considered when dealing with promiscuous mode of operation. The capability
of overhearing the activities of neighbouring nodes within the same wireless transmission range will also enable a node
to get information like nodes identity, source/destination address and other routing/forwarding details. Although the
reason of collecting such information is for a good cause (i.e. to detect selfish node), the privacy of the observed node
may be violated, in which case security issues might arise. Disclosing a nodes private information may cause the
node and the network to be prone to vulnerabilities like stolen identities, denial of service and other illegal acts. The
question is, how do we utilize the advantages of promiscuous listening despite its weaknesses? Hence, researchers
who intend to use promiscuous mode as an observing technique must find acceptable (if not optimal) balance between
those aforementioned aspects such that behaviour information can be obtained at reasonable trade offs. Due to the
issues in passive acknowledgment, other work promote alternative ways to monitor node behaviour by using active
acknowledgment.
In term of routing protocol used to obtain implicit information, source routing such as Dynamic Source Routing
(DSR) has been the most chosen underlying routing protocol by existing cooperation stimulation schemes such as [27,
28, 29, 31, 30, 46]. One reason is due to its built-in optimization of promiscuous listening. With DSR, nodes are able
14

to obtain complete information of an established route, like number of hop counts stored in packets header. However,
other routing protocols such as Ad-hoc On-Demand Distance Vector (AODV) can also use the promiscuous interface
as the integration of the interface with the protocol has been established and is utilized widely. Schemes that rely on
AODV routing protocol and utilized promiscuous mode are such as [41, 42].
4.1.2. Active Acknowledgment
In contrast to the passive acknowledgment approach, active acknowledgment does not rely on MAC layer to
implicitly obtain information on node behaviour. Instead, it uses explicit acknowledgment from peer nodes (either
in end-to-end or hop-by-hop fashion) by transferring acknowledgment (ACK) packet for any successfully received
packets.
End-to-end acknowledgment
In this approach, the destination node is required to send an ACK packet back to the source node to notify that it
has successfully received the packet. If the source node does not receive an expected ACK packet within a specified
time, it will consider the data packet to have failed to reach the destination. In [34], a general framework on reliable
routes and cooperation enforcement policy was proposed to increase performance and reliability of ad hoc networks. A
reliable route was determined based on its reliable index, whereby every detected dropped packet event on a particular
link will reduce the links index. When the reliable index drops beyond the minimum threshold, the route will be
avoided. The dropped packets scenario is detected based on end-to-end acknowledgment whereby the destination
node will send an ACK packet back to the source node upon successfully receiving a packet. A significant weakness
in this work was the assumption of the visibility of every links reliable index to all nodes along the path. The work
in [35] also used the end-to-end acknowledgement method to evaluate the behaviour of neighbouring nodes. In this
scheme, a reputation value is assigned to the next hop neighbour alone, not all nodes along the route. The evaluation
of a node behaviour in [36] was done by relying on TCP acknowledgments. If a packet successfully arrives at the
intended destination, the destination node will send an ACK back to the source. Upon receiving a successful ACK,
each node along the path will increase the reputation index of their neighbouring nodes. Deduction of reputation
index happens when a node realizes that its neighbour did not forward an ACK back or it forwarded a retransmission
of same data packet, which denotes a selfish behaviour.
Hop-by-hop acknowledgment
A similar routine is also applied for the hop-by-hop acknowledgment scheme, except for the length of path that
ACK packets need to travel is shortened, whereby other than destination node, intermediaries are also responsible
for generating and sending ACK packets back to source node. To alleviate the problem of accusing the whole route
consisting of multiple nodes, which causes unnecessary punishment towards innocent nodes, Balakrishnan et al. [37]
proposed a scheme named TWOACK. Unlike any end-to-end acknowledgment scheme, this scheme works by requiring the node located two hops away from the original sender or other forwarder to send an acknowledgment packet
back to the sender upon receiving a data packet. If the waiting-for-ACK-node does not receive any ACK packet
within a specified waiting time, the reputation value of the next hop link decreases and when the value drops to a
certain threshold, the link is marked as misbehaved and would be avoided in the next routing session. This method
helps to detect specific distrusted link rather than punishing the whole route, thus reducing false accusation. In order to reduce traffic congestion due to overwhelming transmission of ACK packets, the authors also introduced an
S-TWOACK (Selective-TWOACK) scheme. The S-TWOACK scheme works by grouping a set of data packets information received from the same upstream link and compile them into one TWOACK packet to be sent back to the
source or forwarder, as a notification of successful packet reception. Both TWOACK and S-TWOACK schemes were
later improved with the introduction of 2ACK scheme proposed by Liu et al. [38]. In the 2ACK scheme, the 2ACK
packet was enhanced with an authentication mechanism using one-way hash chain to preserve the authenticity of the
packet from forging attack. In addition, the routing overhead in the scheme is reasonably low as the acknowledging
process only involves part of data packets instead of all the packets. However, the amount of partially acknowledged
packets must be kept within an acceptable ratio to avoid false report.
Having schemes that utilize acknowledgment for shorter portions of an entire route has helped in reducing the
number of nodes that may be falsely accused as misbehaved. However, such an approach does not outperform the
15

end-to-end acknowledgment technique in detecting a misbehaved node. Even a short link marked as misbehaved
may still contain cooperative nodes. Thus, more misbehaved links need to be detected in order to accurately identify a
misbehaved node. This would require significant evidence gathering to come out with concrete judgment. In addition,
the efficiency of this method is at the expense of high communication overhead due to frequent transmissions of ACK
packets in the network. This is especially true for schemes like 2ACK where every two-hop-away nodes are required
to send an ACK packet back to the sender in order to acknowledge packet reception, thus, increasing the network
traffic that could lead to congestion. Another significant weakness in the schemes that utilize end-to-end or hop-byhop acknowledgment is that false positive is likely to happen such as in the case where destination node does not
generate ACK back to the source or the ACK packet itself is dropped by any node along the way, a misinterpretation
of selfish behaviour could easily occur. In addition, any reliance on TCP ACK makes it unsuitable for transmitting
the packet using UDP, which can lead to higher communication overhead. In order to stimulate cooperation, nodes
should not be burdened with extra forwarding tasks than what they already have.
Probing
Another form of active acknowledgment is based on probing. In this technique, a node monitors another nodes
behaviour by sending out probing packet to any neighbour that is one hop away from the monitored node and expects
an acknowledgment report from that neighbour node. For instance, node A wants to check node Bs forwarding
behaviour, so it needs to send a probing packet to node C which is one hop away from node B. Assuming that node C
is well-behaved and has node Bs forwarding behaviour information, ideally it would send an acknowledgment packet
back to node A to report on node Bs behaviour. Upon receiving the acknowledgment packet, node A gets confirmation
on node Bs forwarding behaviour based on the information sent by node C. Examples of work that have applied the
probing technique are [70, 71]. However, schemes that rely on using packets to probe have some weaknesses, namely,
the dependence on other nodes to verify the behaviour of certain nodes may lead to false judgment in the case where
the node replying to the probing packet gives a false report. In addition, in reality it is hard to be sure of whether
or not a probing packet can be sent to any node that is one hop away from the observed node. In the case where no
one-hop-neighbour of the observed node is visible, there is nowhere to send the probing packet .
4.1.3. Specification-based Approach
Another variance of behaviour observation method is using the specification-based approach, which is a method
of monitoring the behaviour of a certain object and comparing it with a predetermined specification or policy that
the object needs to follow, and check whether or not that object abides by it. Specification-based approach has been
used as an alternative to anomaly detection in identifying behavioural pattern of nodes in wireless multihop networks
whereby focus has been given on creating specifications for routing protocols. In the work by Wang et al. [72],
the authors proposed a specification or policy for the route discovery process in AODV routing protocol that nodes
need to follow by introducing local routing instance (LRI). LRI is a set of ideal transmission processes that happens
during a route discovery session and needs to be completed by a monitored node. In order to keep track of possible
behaviours of every monitored node, the observing node is guided by a finite state machine (FSM) model. The FSM
model acts as a guideline for a local observing node by evaluating sets of possible sequence of behaviours that might
be shown by any monitored node. Through these sequences, node behaviour can be identified based on their ability
to reach the final state of FSM, which is to complete LRI. Selfishness is determined based on number of final states
that a node has failed to reach and has gone beyond certain threshold value. Earlier work on specification-based
approach supported by FSM model was proposed by Tseng et al. [73]. However, this work did not attempt to detect
selfishness, but focused more on malicious behaviour detection launched by attackers. The proposed scheme was the
first to utilize specification-based approach and has been embedded in an intrusion detection system to detect attack
in AODV routing protocol. A similar variant of FSM-based intrusion detection scheme for AODV was proposed by
Ye and Li [74] where classification of several attacks on AODV routing protocol has been made. This technique has
been further applied in many other existing work such as [75, 76].
4.1.4. Game Theory of Perfect/Imperfect Public/Private Monitoring
Apart from being applied in many existing punishment-based mechanisms as discussed earlier, game theory has
also been utilized in observing node behaviour for its capability to provide monitoring analysis model. Several ex16

isting work on detecting node misbehaviour have implemented game theoretical observation components named perfect/imperfect public/private monitoring as alternatives.
Perfect public/private monitoring
Perfect monitoring is a condition where players of a particular initiated game are able to directly observe other
players actions [77, 78]. Subsequently, the entire results (denoted by s ) obtained from the decision making process
by each player are also visible to every other players. Equilibrium is achieved such that no player is known to have
deviated from s to maximize its own payoff which means that every player will always play a fair game. The observed
deviation is the element that can be used to determine consequent actions such as punishment. In the wireless multihop
network context, it is equivalent to a situation where each node is able to observe other nodes behaviour or strategies
at the end of each stage of the game or history of behaviours in previous rounds, in the case of repeated games. This
model is also known as perfect public monitoring where signals (i.e. actions) produced by any node are publicly
observable and shared with other nodes in the same network. In such a case, the signal = (a1 ...an ) and Y where
ai are the actions taken by each player (Y) at every stage of game. In real wireless multihop networks scenarios,
it is not practical to apply the perfect public monitoring concept due to uncertainty nature of node behaviour in the
network where it is difficult to predict that node behaviour is consistent for every round of game. In other words, it
is hard to obtain perfect observatory of node actions where accurate conclusion can be made to perform subsequent
actions. Otherwise, assumption has to be made such that every node has the complete information about other node
behaviours, which is unrealistic.
Imperfect public monitoring
The difficulty of interpreting ambiguous signals can be alleviated using repeated games of imperfect public monitoring [79, 80] in which is a random variable. This method introduces observation based on random public signal
which is correlated with players actions in a game, whereby the signal is assumed to be consistent. In imperfect
public monitoring, players cannot directly observe the actions taken by other players, but may monitor the random
signals produced. The signals can be considered as implicit signs of the actual actions taken by players. For instance,
in tacit collusion under demand fluctuations by Green and Porter [81], two firms that are competing with each other
may not be able to observe the quantities of products that the other firm has produced. However, they can observe the
market price advertised that may reflect the information of product quantities, whereby low price may indicate that
the quantities of production are high. In the wireless multihop network context, this is similar to a scenario where
an observing node may be able to predict other node behaviour based on some indicator of overall network performance such as network goodput. For example, the work done in [82] applied the imperfect monitoring approach to
encourage nodes in ad hoc networks to participate in packet forwarding by observing public signal levels and executing punishments when the level falls below a certain threshold value. In this study, the public signal is indicated
by aggregated network goodput such that if nodes do not deviate from the ideal forwarding behaviour, the signal is
maintained at optimal level. However, if the signal level falls below the minimum threshold that has been set, which
indicates significant node deviation from the game, a trigger point based on Grim trigger strategy is initiated allowing
all nodes to drop packets (i.e. entering punishment phase) causing incentive loss for all involved nodes. Thus, in order
to avoid losing incentive, nodes are self-motivated to keep the signal at the optimum level by not deviating from the
game. Although this kind of strategy seems to work well in encouraging nodes to cooperate, in reality, it is difficult
to obtain information on any selected public signal and disseminate them through the network so that all nodes could
take the signal as reference of the intended behaviours in forwarding packets. In addition, this kind of approach
executes punishment that will affect all nodes in the network, while the real culprit still cannot be detected. Such
punishment method may actually waste network resources due to unnecessary packet dropping by cooperative nodes
when the punishment stage is triggered. The real challenges in applying imperfect public monitoring to stimulate
nodes cooperation in wireless multihop network can be highlighted as follows:
a. Determination of public signal must consist of these properties: (1) it represents the strategies played by each
node; (2) it must be mutually related to the combined strategies of all nodes; and (3) all nodes acknowledge the
signal after each stage of game;
b. Nodes must be willing to unanimously select and share the same random public signal as their reference points;
17

c. Signal must be sufficiently identified in order to execute punishment targeted to deviators; otherwise every node
may suffer from unnecessary punishment;
Considering the above mentioned challenges, it is obvious that the existence of such random public signal is very rare
in wireless multihop networks. Besides, in order to identify the signal, a lot of coordination and information collection
process in the network are required, which could lead to high traffic and communication overhead.
Imperfect private monitoring
In order to counter the above-mentioned issue, another game theoretical monitoring model named imperfect private monitoring [83, 84, 85] was studied. In this model, each action a represents a unique private signal where,

(a1 , . . . , an )

&

Y1

2
..
.

Y2
..
.

Yn

(1)

For this imperfect private monitoring model, the action ai and the signal i are player Yi s private information and
are not observable but may be evaluated based on certain output condition, that is also imperfect. We will discuss an
example application of imperfect private monitoring based on the above-mentioned tacit price cutting. In this case,
price is not observable and the firms sales total is the only signal that can be evaluated but it is private information and
imperfect due to demand fluctuations. However, this information can be used to predict the hidden actions of other
players albeit with a probability of making a wrong decision. Application of imperfect private monitoring method in
wireless multihop networks is still in its infancy stage due to difficulties in maintaining the game recursively and the
lack of information to continue the game for the next stages.
One such approach that has been applied in monitoring the behaviour of nodes in wireless multihop networks is
the work by Ng and Seah [50]. This work adapted a game model that is derived from Aoyagis game of imperfect
private monitoring and communication for Bertrand oligopoly [86]. In general, Aoyagis game is a repeated game
of undisclosed information (product price) and communication (advertisement) among players. Each player would
publicly declare their selling price, but can secretly reduce the published price when dealing with customers, that
would gradually affect the profit gained by other players. The problem lies in how to observe the privately released
signals so that appropriate actions can be taken by other players to avoid further loss. In order to solve the problem,
the author in [86] introduced a collusive communication among players where each player chooses a collusive price
(pi ) and eventually reveal their private signals at the end of the collusion stage. Similar communication element has
been implemented in [87, 88]. The private signal d is reported in the form of either 0 or 1 indicating low or high
signal respectively that is determined by comparing with threshold value (mi ) . The decision on whether or not the
communication remains in collusive mode or turn into punishment mode is highly dependent on the uniformity of
the reported private signals by every player. As opposed to imperfect public monitoring where the observation signal
obtained does not necessarily reflect the real action of each player, imperfect private monitoring on the other hand
does provide more accurate signal. However, it is difficult to implement as there is still no guarantee of obtaining a
genuine signal. Rationally, each player would still possess a likelihood to lie (i.e. give false report) but equilibrium
is built in such a way that incentive will be given when genuine reporting is maintained or else punishment will take
place.
4.2. Node Behaviour Judgment
Judging a nodes intentions by observing its behaviour can be very challenging especially in a wireless network,
where information can be lost or altered by the environment. In this section, we examine the various techniques used
to assess a nodes behaviour so that the outcome can be used to determine whether a node is cooperative or not.
4.2.1. Local vs. Global Information
In order to gain the information on node behaviour, it is important to identify the source of behaviour information
i.e. whether it should be obtained locally or globally. Local or direct observation means that a node evaluates the
18

behaviour of other nodes based on its personal experiences and observations. In contrast, getting information globally
(i.e. indirect observation) implies that a node gains information on a particular node behaviour based on the external
opinions of other nodes to support its evaluation. In this case, information dissemination will most likely occur in
the network in order for nodes to exchange information. As shown in Table 3, analysis on information source shows
that most payment-based schemes do not apply any behaviour observation methods, unlike reputation-based and
punishment-based schemes, due to the nature of these schemes which rely on credit system whereby, loss of credits is
regarded as an automatic punishment to selfish nodes.
Local observation is applied in most existing incentive-based mechanism and is considered as a compulsory procedure due to the important need for every node to self-evaluate other nodes prior to getting external information. The
difference is in terms of whether or not local observation is exchanged with other nodes in the network and obtained
information is supported by global reports. Schemes that rely on local observations only, such as [31], were motivated
by the fact that issues in getting global information such as false accusation, unscalable information management
and unnecessary overhead due to information flooding are avoided. However, this option is applied at the expense of
getting insufficient information about the behaviour of other nodes. On the other hand, nodes that obtain behaviour
information based on reports from other nodes are prone to risk of getting false reports, especially when the other
nodes collude with each other to mask the real behaviour of a particular node for self-benefits. Hence, the decision of
determining the information provider is critical in any incentive-based mechanism for detecting the real behaviour of
a particular observed node.
Differences between implementing local and global observation in determining node behaviour in wireless multihop networks are summarized in Table 2. Based on the summary, there are pros and cons in determining the types
Table 2: Local Observation vs. Global Observation
Sources
Elements

Local Observation

Topology

Could adapt well

changes of topology

Overhead

Less communication overhead

Scalability
Information
Sufficiency
Node
Collusion

Global Observation
with

dynamic

More scalable due to behaviour information is viewed from the perspective

of a single node and need not be shared
or exchanged
May not have sufficient information on
the overall state of node behaviour
No risk of node collusion but false accusation may occur

More suitable for static topology

More communication overhead due to
information exchange/distribution
Less scalable due to the need for information to be shared across network and
to maintain global behaviour information of other nodes
Able to collect multiple opinions from
other nodes
Higher risk of getting false accusation
by colluding nodes

of observation sources between local and global observation for node behaviour evaluation. The choice depends on
the efficiency of each method in providing information on selfishness at reasonable trade-offs among elements listed
in the table. However, until now, each method still needs to be improved to strengthen the authenticity of behavioural
information that are observed and possibly shared across the network.
4.2.2. Monitored Behaviour
At the network layer, two major node functions are routing and forwarding data packets which can be observed
and evaluated by an observer to determine cooperativeness/selfishness behaviour. The effort in forwarding packets
can be measured using performance metrics like forwarding rate and number of dropped packets. Other factors like
dropping probability and wireless transmission error might also be considered to get a more accurate measurement as
has been applied in [45, 46]. With that implies, a threshold value is set to determine the level of nodes effort as being
either cooperative or selfish.
19

4.3. Information Broadcast

The obtained behaviour information may need to be shared with other nodes to strengthen judgment and get
concrete evaluation. The class of reputation-based schemes is among ones that require behaviour information to be
broadcast in order to determine the reputation level of a node based on votes from other nodes. This would require
efficient techniques to disseminate the information so that it reaches intended destinations correctly. One important
aspect is on the underlying routing protocol used. With respect to information broadcast, DSR has more advantage
due to the fact that a source node knows the information of other nodes in the entire established route. In addition,
nodes in DSR area able to learn and store multiple routes. This means, behaviour information can be shared with
large number of known nodes without having to flood and exchange the packets containing such information to
anonymous nodes. Unlike DSR, nodes in AODV only know their next hop neighbours. Thus, it has to depend on
other nodes to broadcast information to nodes of beyond its channel coverage. Although storage overhead can be
reduced, broadcasting information using AODV would cause higher communication overhead. Nonetheless, since
both are on-demand protocol, having to flood packets blindly to exchange reputation information is unavoidable.
Thus, a pro-active routing approach might be a worthwhile consideration.
In [89], broadcasting and exchanging reputation information have been evaluated based on both on-demand and
pro-active techniques. Both techniques were found to have advantages and drawbacks in handling such processes.
Although on-demand technique has less communication overhead, but with its establishment being temporary, some
information might not be shared in timely fashion. Whereas, pro-active technique is able to broadcast information at
any time due to its steady-state readiness, but at the cost of having high network overhead.
4.4. Summary
A comparison of the various cooperation stimulation mechanisms was carried out and the results are presented in
Table 3. We identified the pros and cons of each element, together with general strength and flaws of each category of
schemes.
5. Open Research Issues
Following the review encompassing essential elements in cooperation stimulation mechanisms, we will further
discuss several related issues that are still open research problems in this field. The main highlighted issues are false
judgment and node collusion, whereby these problems originated from the inaccuracy in the evaluation of a particular
nodes behaviour.
5.1. False Judgment
As aforementioned, the most basic procedure to obtain information on behaviour is through the observation of a
single node, where the information may be shared with other nodes or kept for self-reference. The accuracy of the
information is susceptible to false judgment especially when a node behaviour is judged based on single set of actions
(SSA) using metrics like number of dropped packets and forwarding rate. SSA is a condition of when a node observes
another nodes forwarding effort towards its own requests while disregarding that nodes effort towards other nodes
requests [90]. This procedure is applied in most of the existing work, whereby the scenario that is most likely to occur
due to this practice is that any packet dropping activity (i.e. above certain threshold) by the observed node can be
regarded as selfish, which may not be assessed sufficiently. This means that it is not always fair to evaluate a nodes
effort based on SSA since the observed node may have constraints in fulfilling the request due to other problems, like
having to serve heavy incoming traffic.
Apart from having the fairness issue, the lack of adequate information as a consequence of the SSA approach is
also a key factor in the reliance on global reports. Commonly, this process has to be done in order to gain assurance
as to whether or not the same observed node exudes the same behaviour, as evaluated by other nodes in the network,
based on the number of majority votes. However, since the global report is formed based on the information collected
from a group of observing nodes that also utilizes SSA, the information may not accurately reflect a particular node
actual behaviour, which leads to false judgment. Having said that, the problem of false judgment actually has a cyclic
effect, as shown in Figure 4. When a local node requests for global report which is based on the majority of votes
cast by other local nodes, the authenticity of the behavioural information may already be contaminated by several
20

Table 3: Summary of Comparison

Type

Monitoring
Mechanism

Information Source
(Local VS Global)

Punishment Scheme

General Strength and Flaws

Strength
- Credits as incentive motivating
node to cooperate
Flaws
- Mostly, selfish node is not
detected
- High communication and
processing overhead
- Bottleneck
- Storage overhead
- Unfair share of credits
distribution due to node
location
- Reliance on costly tamper-proof
device

Not Available
[14-19][21-26]
Pros:
- No monitoring overhead
- Less energy consumption
Cons:
Exact selfish node is not
detected
End-to-end
ACK [20]

Reputation
Scheme (RS)
Watchdog[27]
CONFIDANT[28]
CORE [29]
SORI[30]
OCEAN[31]
HEAD[32]
LARS[33]
Performability[34]
Malicious
Counter[35]
Node Isolation[36]
TWOACK and STWOACK [37]
2ACK[38]
SDA[39]
TEAM[41]
Trust-based[42]
DARWIN[46]

Promiscuous Mode
[27][28][29][30][31][32]
[33][39][41][42][46]

Local Observation
[27][28][29][30][31][32]
[33][34][35][36][37][38]
[39][41][42][46]

Pros:
- Self-evaluation
- Less false report, unless
intentionally faked
Cons:
- Unreliable channel
- High energy
consumption
- Privacy issue
End-to-end ACK
[34][35][36]
Pros:
- Not rely on MAC layer
- Less ACK packets
propagation
Cons:
- Punishing whole link
- Less false report, unless
intentionally faked
Hop-by-hop ACK
[37][38]
Pros:
- Detect specific distrusted
link
Cons:
- High communication
overhead

Pros:
- Scalable
- Less overhead
- Adapt with dynamic
topology changes
Cons:
- False judgment
- Insufficient information

Pros:
Possible second chance
Cons:
Avoided selfish node can
get away from network
tasks.

Global Observation
[28][29][30][39][41][42]
Pros:
- Strengthen local
judgment
Cons:
- Node collusion
- High communication
overhead
- Susceptible to false
accusation
- Less scalable

Mass Enforcement
(Path deletion, denial of
service request, node
isolation/ blacklisted,
denial of access control)
[28][29][30][31][33][34]
[35][36][39][41][46]

Promiscuous Mode
[47][48]

Local Observation
[38][39][48]

End-to-end ACK
[45]

Global Observation
[38][39][48]

Pros:
- Punishment motivates
node to cooperate in
next stage
Cons:
- No redemption may
cause resource wastage
- Possible false
accusation
Mass Enforcement
(Denial of service request,
node isolation, denial of
access control)
[38][39][48]

Pros and cons similar to

RS.

Pros and cons similar to

RS.

Pros and cons similar to

RS.

Hybrid Scheme
(HS)
ICARUS[45]
ARM[47]
RCM[48]

Not Available
[14][16][17][18][19]
- Nodes are not bounded
to observation task to
detect selfish node.
- Mainly rely on credit
transactions to motivate
cooperative behaviour.

Not Enforced
[14-18][21-23][25-26]
Pros:
Less processing overhead
Cons:
Undetected selfish node
can get away from
network tasks.
Mass Enforcement
Denial of service [19][24]
Certificate retraction [20]
Pros:
Node vindication
Cons:
False judgment
periphery node less
chance to collect credit
Mild Enforcement
(Path avoidance)
[27][32][37][38][42]

Credit-Payment
Scheme (CPS)
Nuglets [14,15,16]
Sprite[17]
Adhoc VCG[18]
COMMIT [19]
Express[20]
RACE[21]
Micro-Payment[22]
Charge-Reward[23]
TRIPO[24]
FESCIM[25]
CASHnet[26]

Pros:
Verification of nodes
effort is strengthened
Cons:
Communication overhead

21

Strength
- Motivate node to increase
reputation level by being
cooperative
- Strengthened behaviour
evaluation by using voting
mechanism
- Unsusceptible to false judgment
by small group of cheaters
Flaws
- Prone to false accusation by
large group of colluding nodes
- Complex trust management
design in large topology
- High resource consumption and
communication overhead

Strength
- Credits as incentive motivating
node to cooperate
- More balanced charge and
reward amount
Flaws
- Possibly very high
communication and processing
overhead due to the hybrid
features

Requesting Node
Local
Observation
(SSA)
REQUEST FOR
GLOBAL REPORT

Local
Observation
(SSA)

GLOBAL REPORT
PROVISION
[Possible
false judgment]
Observation report
[Possible
false judgment]

Data to reinforce judgment

based on local observation

Observation report
[Possible
false judgment]

Data to reinforce judgment

based on local observation

Data to reinforce judgment

based on local observation

Local
Observation
(SSA)

Local
Observation
(SSA)

Observation report
[Possible flawed/ bias/
tampered judgment]

Data to reinforce judgment

based on local observation

Observation report
[Possible
Data to reinforce judgment false judgment]
based on local observation

Local
Observation
(SSA)

Observation report
[Possible
accurate judgment]

Global Observation
Report
(Majority of votes)

Local
Observation
(SSA)

Observation report
[Possible flawed/ bias/
tampered judgment]

Data to reinforce judgment

based on local observation

[Possible node
collusion]

Local
Observation
(SSA)

Figure 4: The Circulating Problem of False Judgment based on SSA

factors, such as, biased and tampered judgment. Even with the increasing numbers of voters, a node may only obtain
information on selfishness or cooperativeness based on highest portion of votes, while the false judgment problem
may still be circulating within the loop. This is because every single observing node has the authority to judge a
particular node behaviour based on their own stochastic perceptions by using SSA. Thus, the SSA approach needs to
be reassessed where the reliability of the observed information by every single node must be guided by a more efficient
mechanism based on policies and/or agreed criteria. The element that needs to be highlighted and focused here is the
need to obtain the information on how cooperative and fairly a nodes effort is towards different arriving requests, as
promptly and accurately as possible. Thus, the aim is to reduce the necessity of having to rely on the global report, as
much as possible. As matter of fact, having more accurate local observation information will strengthen the accuracy
of the global report if, at certain point, it is deemed necessary by any particular node.
Hence, the accuracy of the common observation routine must be assured, so that an observer node has a smaller
chance of making a false judgment. One of the consequences of a false judgment is its effect on global sharing
accuracy when the shared information is flawed. Although some reputation mechanisms such as [31] introduced
rebound state element by giving a second chance (i.e. redemption period) to a falsely accused node to re-participate
in the network, it is more desirable if the reliability of the first-hand information can be guaranteed beforehand. As
shown in Figure 3, several elements of the cooperation stimulation mechanism following the behaviour judgment part
may be adversely affected as a consequence of having false judgment, and may result in a defective overall cooperation
stimulation mechanism. Hence, the behaviour information obtained must be strengthened with some mechanisms that
could portray the viability of the reported behaviour which will be further investigated in our study.
5.2. Node Collusion
Node collusion is another issue that is highlighted in this article as depicted in Figure 4. It is not wrong for a group
of nodes in a wireless multihop networks to agree on the same behavioural information on a particular node. However,
it is not right for them to collude by falsely accusing or praising any node. When this scenario happens, a good node
may be unnecessarily punished while a bad node remains undetected in the network. This is a problem that has yet
to be addressed successfully in the sense that there is no mechanism to control the reliability of information that the
nodes decide to share. Any node requesting information on a particular nodes reputation, whether good or bad, would
22

have to rely on the majority votes. Lack of an accurate method to judge a node behaviour makes it convenient for any
node to declare false information because assessment is not guided by any formal metrics or policy.
Intuitively, if nodes are bounded by certain metrics when judging other nodes, where they need to present certain
evidences to prove that fair evaluation has been made, the risk of collusion could be reduced. Hence, it is important
to provide a mechanism that can ensure the accuracy and reliability of a node behaviour such as a recently proposed
method called compare and measure selfishness detection [90]. Above all, the question will return to what type
of monitoring techniques should be used in order to obtain accurate information. Given the discussion on the pros
and cons of currently available monitoring techniques in Section 4.1, thoughtful consideration must be made prior to
making any selection.
5.3. Accuracy
The accuracy of behaviour information is the most fundamental element that has not been fully addressed in
existing schemes, which is why the issue of false judgment is still affecting the efficiency of the proposed schemes.
Although there exist proposals to improve accuracy from the perspective of physical, MAC and network layers, most
come with the assumption that information can be sufficiently collected by using the SSA approach. Even worse,
some simply assume that selfish behaviour can be assessed without explicitly stating how such information is actually
measured. Hence, deeper investigation needs to be done to improve existing approaches especially SSA, that could
provide faster, yet more conclusive behaviour information.
5.4. Scalability
To improve the scalability of cooperation mechanism, it is good to reduce the dependency on centralized approach
in managing nodes in the networks. Although having any node appointed as central agent to function as a trust
manager or credit administrator would help in coordinating some processes smoothly, but as far as wireless multihop
networks characteristics is concerned, it would introduce delays, bottlenecks and overhead when every other nodes
need to communicate and submit reports to the agent. Having such dependancy, the more nodes joining the networks,
the more hectic the networks would be, leading to scalability issues. However, having distributed approaches also
come with some drawbacks as every node may have the authority to make autonomous decision. Hence, nodes that
work under distributed mode must be guided with a set of policies, though not as stringent as having to rely on a
central agent. Having said that, further investigation on how to exploit the advantageous elements of both distributed
and centralized approaches must be carried out.
5.5. Security
Security and cooperation are cross layers issues in the detection of node misbehaviour in networks. While cooperation approaches aim to address the selfishness issue, security approaches on the other hand aim to deal with malicious
behaviour. Although selfishness is a subset of malicious behaviour, security mechanisms targeting malicious behaviour may not necessarily be suitable countermeasures for selfishness. But, it cannot be denied that security aspects
must also be considered in the design of cooperation stimulation mechanisms, such as, preserving node privacy during
monitoring, having authenticated acknowledgment packets and guarantee on non-spoofing of node identity. Hence,
there is a need of providing optimal balance between security and cooperation approaches such that elements from
both approaches would complement each other to produce the desired outcomes.
5.6. Application
Most existing cooperation stimulation mechanisms have been proposed for general wireless multihop networks
but not tailored to meet the specialized requirements of any particular network type. For example, credit-payment
scheme with central agent may be more suitable for WSNs where nodes are less mobile, as compared to applications
on MANETs where node mobility is high and appointing a central agent may not be feasible for the ad hoc nature
of the networks. Thus, there is a need to design a cooperation mechanism that would specifically cater to unique
characteristics of different network types.

23

6. Conclusion
This article presents an in-depth survey on cooperation in wireless multihop networks focusing on cooperation
stimulation mechanisms that have been proposed to address the issue of selfishness. Several aspects in the proposed
mechanisms like the adopted monitoring and selfishness detection approaches have been discussed in order to identify possible weaknesses prior to finding ways for addressing them. These components are among those presented
in the comparative analysis table (Table 3), which play important roles in the development of any cooperation stimulation mechanism. The comparison of the various schemes led to the highlighted issues that will motivate research
to find ways in providing accurate measurement of a node behaviour. Having more accurate behaviour measurement
is expected to address issues like false judgment and node collusion in wireless multihop networks. To date, these
aforementioned issues are open research problems in this area, where proposed solutions still have room for improvement. In view of these challenges, cooperation stimulation mechanisms in wireless multihop networks still require
extensive investigations and rigorous research studies, providing abundant opportunities for novel discoveries and new
innovations.
References
[1] Yick, J., Mukherjee, B., Ghosal, D.: Wireless sensor network survey. Computer Networks 52(12), 2292-2330 (2008).
[2] Sharef, B. T., Alsaqour, R. A., Ismail, M.: Vehicular communication ad hoc routing protocols: A survey. Journal of Network and Computer
Applications 40, 363-396 (2014).
[3] Komali, R. S., MacKenzie, A. B. and Gilles, R. P. : Effect of selfish node behavior on efficient topology design. IEEE Transactions on Mobile
Computing 7(9), 1057-1070 (2008).
[4] Yoo, Y., Agrawal, D.P. : Why does it pay to be selfish in a MANET?. IEEE Wireless Communications 13(6), 87-97 (2006).
[5] Sanzgiri, K., LaFlamme, D., Dahill, B., Levine, B. N., Shields, C., Belding-Royer, E. M.: Authenticated routing for ad hoc networks. IEEE
Journal on Selected Area in Communications 23(3), 598-610 (2005).
[6] Zhong, S., Li, L. E., Liu, Y. G., Yang, Y. R. : On designing incentive-compatible routing and forwarding protocols in wireless ad-hoc
networks: an integrated approach using game theoretic and cryptographic techniques. Wireless Networks 13(6), 799-816 (2007).
[7] Myerson, R. B.: Game Theory: Analysis of Conflict. Harvard University Press, 1991.
[8] Herhold, P., Zimmermann, E., Fettweis, G.: Cooperative multi-hop transmission in wireless networks. Computer Networks 49(3), 299-324
(2005).
[9] Sharma, S., Yi, S., Hou, Y. T., Sherali, H. D., Kompella, S.: Cooperative communications in multi-hop wireless networks: joint flow routing
and relay node assignment. In: Proceedings of the 29th IEEE International Conference on Information Communications (INFOCOM), San
Diego, CA, 14-19 March 2010, pp. 1-9.
[10] Sharma, S., Yi, S., Hou, Y. T., Kompella, S.: An optimal algorithm for relay node assignment in cooperative ad hoc networks. IEEE/ACM
Transactions on Networking 19(3), 879-892 (2011).
[11] Zhang, J., Zhang, Q. : Cooperative routing in multi-source multi-destination multi-hop wireless networks. In: Proceedings of the 27th IEEE
International Conference on Information Communications (INFOCOM), Phoenix, AZ, 13-18 April 2008, pp. 1-5.
[12] Lin, Y., Song, J.-H., Wong, V.: Cooperative protocols design for wireless ad-hoc networks with multi-hop routing. Mobile Networks and
Applications 14(2), 143-153 (2009).
[13] Khandani, A.E., Abounadi, J., Modiano, E., Lizhong, Z. : Cooperative Routing in Static Wireless Networks. IEEE Transactions on Communications 55(11), 2185-2192 (2007).
[14] Buttyan, L., Hubaux, J.-P.: Nuglets: A virtual currency to stimulate cooperation in self-organized mobile ad-hoc networks. In: Technical
Report DSC/2001/001. Swiss Federal Institute of Technology, Lausanne, Switzerland, (2001).
[15] Buttyan, L., Hubaux, J. P.: Stimulating cooperation in self-organizing mobile ad hoc networks. Mobile Networks and Applications 8(15),
579-592 (2002).
[16] Buttyan, L., Hubaux, J.-P.: Enforcing service availability in mobile ad-hoc WANs. In: Proceedings of the 1st IEEE/ACM Workshop on
Mobile and Ad Hoc Networking and Computing (MobiHOC), Boston, MA, 11 August 2000, pp. 87-96.
[17] Zhong, S., Yang, Y. R., Chen, J.: Sprite: A simple, cheat proof, credit-based system for mobile ad hoc networks. In: Proceedings of the 22nd
IEEE International Conference on Information Communications (INFOCOM), San Francisco, USA, 1-3 April 2003, pp. 1987-1997.
[18] Anderegg, L., Eidenbenz, S. : Ad hoc-VCG: A truthful and cost-efficient routing protocol for mobile ad hoc networks with selfish agents. In:
Proceedings of the 9th International Conference on Mobile Computing and Networking (MobiCom), San Diego, CA, 14-19 September 2003,
pp. 245-259.
[19] Eidenbenz, S., Resta, G., Santi, P.: COMMIT: A sender-centric truthful and energy-efficient routing protocol for ad hoc networks with selfish
nodes. In: Proceedings of 19th IEEE International, Parallel and Distributed Processing Symposium (IPDPS), Denver, Colorado, USA, 3-8
April 2005, pp. 239-249.
[20] Janzadeh, H., Fayazbakhsh, K., Dehghan, M., Fallah, M. S.: A secure credit-based cooperation stimulating mechanism for MANETs using
hash chains. Future Generation Computer Systems 25(8), 926-934 (2009).
[21] Mahmoud, M. M. E. A., Shen, X.(S).: A secure payment scheme with low communication and processing overhead for multihop wireless
networks. IEEE Transactions on Parallel and Distributed Systems 24(2), 209-224 (2013).
[22] Jakobsson, M., Buttyan, L., Hubaux, J.-P.: A micro-payment scheme encouraging collaboration in multi-hop cellular networks. In: Proceedings of 7th International Financial Cryptography Conference, Gosier, Guadeloupe, 27-30 January 2003, pp. 15-33.

24

[23] Salem, N. B., Buttyan, L., Hubaux, J.-P., Jakobsson, M.: Node cooperation in hybrid ad hoc networks. IEEE Transactions on Mobile
Computing, 5(4), 365-376, 2006.
[24] Mahmoud, M. M. E. A., Shen, X.(S).: An integrated stimulation and punishment mechanism for thwarting packet drop in multihop wireless
networks. IEEE Transactions on Vehicular Technology 60(8), 3947-3962 (2011).
[25] Mahmoud, M. M. E. A., Shen, X.(S).: FESCIM: Fair, Efficient, and Secure Cooperation Incentive Mechanism for hybrid ad hoc networks.
IEEE Transactions on Mobile Computing 11(5), 753-766 (2012).
[26] Weyland, A., Staub, T., Braun, T.: Comparison of motivation-based cooperation mechanisms for hybrid wireless networks. Computer Communications 29(13-14), 2661-2670 (2006).
[27] Marti, S., Giuli, T. J., Lai, K., Baker, M. : Mitigating routing misbehavior in mobile ad hoc networks. In: Proceedings of the 6th Annual
International Conference on Mobile Computing and Networking, Boston, Massachusetts, USA, 6-11 August 2000, pp. 255-265.
[28] Buchegger, S., Boudec, J.-Y. L. : Performance analysis of the CONFIDANT protocol. In: Proceedings of 3rd ACM International Symposium
on Mobile Ad hoc Networking and Computing (MOBIHOC), Lausanne, Switzerland, 9-11 June 2002, pp. 226-236.
[29] Michiardi, P., Molva, R.: CORE: A collaborative reputation mechanism to enforce node cooperation in mobile ad hoc networks. In: Proceedings of 6th Joint Working Conference on Communications and Multimedia Security, Netherlands, 26-27 September 2002, pp. 107-121.
[30] He, Q., Wu, D., Khosla, P.: SORI: A secure and objective reputation based incentive scheme for ad hoc networks. In: Proceedings of IEEE
Wireless Communications and Networking Conference (WCNC), New Orleans, LA, USA, 21-25 March 2004, pp. 825-830.
[31] Bansal, S., Baker, M.: Observation-based cooperation enforcement in ad hoc networks. In: Technical Report cs.NI/0307012. Computer
Science Department, Stanford University, USA, (2003).
[32] Guo, J., Liu, H., Dong, J., Yang, X.: HEAD: A hybrid mechanism to enforce node cooperation in mobile ad hoc networks. Tsinghua Science
and Technology 12(1), 202207 (2007).
[33] Hu, J., Burmester, M.: LARS: A locally aware reputation system for mobile ad hoc networks. In: Proceedings of the 44th Annual Southeast
Regional Conference, Melbourne, Florida, USA, 10-12 March 2006, pp. 119-123.
[34] Conti, M., Gregori, E., Maselli, G.: Towards reliable forwarding for ad hoc networks. In: Proceedings of Personal Wireless Communications
(PWC), Venice, Italy, 23-25 September 2003, pp. 790804.
[35] Dewan, P., Dasgupta, P., Bhattacharya, A.: On using reputations in ad hoc networks to counter malicious nodes. In: Proceedings of the 10th
International Conference on Parallel and Distributed Systems (ICPADS), Newport Beach, CA, USA, 7-9 July 2004, pp. 665-672.
[36] Refaei, M.T., Srivastava, V., DaSilva, L., Eltoweissy, M.: A reputation-based mechanism for isolating selfish nodes in ad hoc networks. In:
Proceedings of Second Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services (MobiQuitous), San
Diego, California, USA, 17-21 July 2005, pp. 3-11.
[37] Balakrishnan, K., Deng, J., Varshney, P.K.: TWOACK: Preventing selfishness in mobile ad hoc networks. In: Proceedings of IEEE Wireless
Communications and Networking Conference (WCNC), New Orleans, LA, USA, 13-17 March 2005, pp. 2137-2142.
[38] Liu, K., Deng, J., Varshney, P.K., Balakrishnan, K. : An acknowledgment-based approach for the detection of routing misbehavior in
MANETs. IEEE Transactions on Mobile Computing 6(5), 536550 (2007).
[39] Chong, Z.-K., Tan, S.-W., Goi, B.-M., Ng, B.C.-K.: Outwitting smart selfish nodes in wireless mesh networks. International Journal of
Communication Systems 26(9), 11631175 (2013).
[40] Buchegger, S., Le Boudec, J-Y.: The effect of rumor spreading in reputation systems for mobile ad-hoc networks. In: Proceedings of the
Workshop on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks (WiOpt), INRIA Sophia-Antipolis, France, 3-5 March
2003, pp. 1-10.
[41] Ferraz, L. H. G., Velloso, P. B., Duarte, O. C. M. B.: An accurate and precise malicious node exclusion mechanism for ad hoc networks. Ad
Hoc Networks 19, 142-155 (2014).
[42] Eissa, T., Abdul Razak, S., Khokhar, R., Samian, N.: Trust-Based routing mechanism in MANET: design and implementation. Mobile
Networks and Applications 18(5), 666-677 (2013).
[43] Yu, M., Zhou, M., Su, W.: A secure routing protocol against byzantine attacks for manets in adversarial environments. IEEE Transactions on
Vehicular Technology 58(1), 449-460 (2009).
[44] Zhang, C., Zhu, X., Song, Y., Fang, Y.: A formal study of trust-based routing in wireless ad hoc networks. In: Proceedings of the 29th IEEE
International Conference on Information Communications (INFOCOM), San Diego, CA, USA, 15-19 March 2010, pp. 1-9.
[45] Charilas, D.E., Georgilakis, K. D., Panagopoulos, A. D.: ICARUS: hybrId inCentive mechAnism for coopeRation stimUlation in ad hoc
networkS. Ad Hoc Networks 10(6), 976-989 (2012).
[46] Jaramillo, J.J., Srikant, R.: DARWIN: distributed and adaptive reputation mechanism for wireless ad-hoc networks. In: Proceedings of the
13th annual ACM international conference on Mobile computing and networking (MobiCom 07), Montreal, Canada, 9-14 September 2007,
pp. 87-98.
[47] Shen, H., Li, Z.: ARM: An account-based hierarchical reputation management system for wireless ad hoc networks. In: 28th International
Conference on Distributed Computing Systems Workshops (ICDCS 08), Beijing, China, 17-20 June 2008, pp. 370-375.
[48] Wang, X., Cai, Y., Li, Z.: A novel hybrid incentive mechanism for node cooperation in mobile cyber-physical systems. International Journal
of Parallel, Emergent and Distributed Systems 29(3), 316-336 (2014).
[49] Milan, F., Jaramillo, J. J., Srikant, R.: Achieving cooperation in multihop wireless networks of selfish nodes. In: Proceedings of Workshop
on Game Theory for Communications and Networks (GameNets), Pisa, Italy, 14th October 2006, pp. 1-10.
[50] Ng, S.K., Seah, W. K. G.: Game-theoretic approach for improving cooperation in wireless multihop networks. IEEE Transactions on Systems,
Man, and Cybernetics, Part B: Cybernetics 40(3), 559-574 (2010).
[51] Felegyhazi, M., Hubaux, J. P., Buttyan, L.: Nash equilibria of packet forwarding strategies in wireless ad hoc networks. IEEE Transactions
on Mobile Computing 5(5), 463-476 (2006).
[52] Niu, B., Zhao, H. V., Hai, J.: A cooperation stimulation strategy in wireless multicast networks. IEEE Transactions on Signal Processing
59(5), 2355-2369 (2011).
[53] Han, Z., Ji, Z., Liu, K. J. R.: A cartel maintenance framework to enforce cooperation in wireless networks with selfish users. IEEE Transactions on Wireless Communications 7(5), 18891899 (2008).

25

[54] Fudenberg, D., Tirole, J.: Game Theory. MIT Press, Cambridge, MA (1991)
[55] Srivastava, V., Neel, J., Mackenzie, A. B., Menon, R., Dasilva, L. A., Hicks, J. E., Reed, J. H., Gilles, R. P.: Using game theory to analyze
wireless ad hoc networks. IEEE Communications Surveys and Tutorials 7(4), 46-56 (2005).
[56] Akkarajitsakul, K., Hossain, E., Niyato, D., Kim, D. I.: Game theoretic approaches for multiple access in wireless networks: A survey. IEEE
Communications Surveys and Tutorials 13(3), 372-395 (2011).
[57] MacKenzie, A.B., DaSilva, L.: Game theory for wireless engineers. Morgan and Claypool. (2006).
[58] Charilas, D. E., Panagopoulos, A. D.: A survey on game theory applications in wireless networks. Computer Networks 54(18), 3421-3430
(2010).
[59] Huang, L., Li, L., Liu, L., Zhang, H., Tang, L.: Stimulating cooperation in route discovery of ad hoc networks. In: Proceedings of the 3rd
ACM workshop on QoS and security for wireless and mobile networks (Q2SWinet 07), Chania, Crete Island, Greece, 22-26 October 2007,
pp. 39-46.
[60] Zhang, Z., Shi, J., Chen, H-H., Guizani, M., Qiu, P.: A cooperation strategy based on nash bargaining solution in cooperative relay networks.
IEEE Transactions on Vehicular Technology 57(4), 2570-2577 (2008).
[61] Sun, Y., Guo, Y., Ge, Y., Lu, S., Zhou, J., Dutkiewicz, E.: Improving the transmission efficiency by considering non-cooperation in ad hoc
networks. The Computer Journal 56(8), 1034-1042 (2013).
[62] Lee, J.-F., Liao, W., Chen, M. C.: An incentive-based fairness mechanism for multi-hop wireless backhaul networks with selfish nodes. IEEE
Transactions on Wireless Communications 7(2), 697-704 (2008).
[63] Zhong, S., Wu, F.: A collusion-resistant routing scheme for noncooperative wireless ad hoc networks. IEEE/ACM Transactions on Networking 18(2), 582-595 (2010).
[64] Refaei, M. T., Rong, Y., DaSilva L. A, Choi, H-A.: Detecting node misbehavior in ad hoc networks. In: Proceedings of IEEE International
Conference on Communications (ICC), Glasgow, Scotland, 24-28 June 2007, pp. 3425-3430.
[65] Toledo, A. L., Wang, X.: Robust detection of selfish misbehavior in wireless networks. IEEE Journal on Selected Areas in Communications
25(6), 1124-1134 (2007).
[66] IEEE Standard for Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, Nov. 1997. P802.11.
[67] Lei, H., Liu, L.: Extended watchdog mechanism for wireless sensor networks. Journal of Information and Computing Science 3(1), 39-48
(2008).
[68] Liang, G., Agarwal, R., Vaidya, N.: When watchdog meets coding. In: Proceedings of the 29th IEEE International Conference on Information
Communications (INFOCOM), San Diego, CA, USA, 15-19 March 2010, pp. 2267-2275.
[69] Hernandez-Orallo, E., Serrat, M. D., Cano, J., Calafate, C. T., Manzoni, P.: Improving selfish node detection in MANETs using a collaborative
watchdog. IEEE Communications Letters 16(5), 642-645 (2012).
[70] Just, M., Kranakis, E., Wan, T.: Resisting malicious packet dropping in wireless ad hoc networks. In: Proceedings of International Conference
on Ad-Hoc, Mobile, and Wireless Networks (ADHOC-NOW), Montreal, Canada, 8-10 October 2003, pp. 151-163.
[71] Awerbuch, B., Holmer, D., Nita-Rotaru, C., Rubens, H.: An on-demand secure routing protocol resilient to byzantine failures. In: Proceedings
of the ACM Workshop on Wireless Security (WiSE), Atlanta, GA, USA, 28 Sept. 2002, pp. 21-30.
[72] Wang, B., Soltani, S., Shapiro, J. K., Tan, P.-N.,: Local detection of selfish routing behavior in ad hoc networks. In: Proceedings of 8th
International Symposium on Parallel Architectures, Algorithms and Networks (ISPAN), Las Vegas, Nevada, USA, 7-9 Dec. 2005, pp. 392399.
[73] Tseng, C.-Y., Balasubramanyam, P., Ko, C., Limprasittiporn, R., Rowe, J., Levitt, K.: A specification-based intrusion detection system for
AODV. In: Proceedings of the 1st ACM Workshop on Security of Ad hoc and Sensor Networks, Fairfax, Virginia, 27-30 Oct. 2003, pp.
125-134.
[74] Ye, X., Li, J.: An FSM-based automatic detection in AODV for ad hoc network. In: Proceedings of International Symposium on Computer
Network and Multimedia Technology (CNMT), Wuhan, China, 18-20 Dec. 2009, pp. 1-5.
[75] Orset, J.-M., Alcalde, B., Cavalli, A.: An EFSM-based intrusion detection system for ad hoc networks. In: Proceedings of the 3rd Automated
Technology for Verification and Analysis (ATVA), Taipei, Taiwan, 4-7 Oct. 2005, pp. 400-413.
[76] Lin, H.-C., Sun, M.-K., Huang, H.-W., Tseng, C.-Y.H., Lin, H.-T.: A specification-based intrusion detection model for wireless ad hoc networks In: Proceedings of the 3rd International Conference on Innovations in Bio-Inspired Computing and Applications (IBICA), Kaohsiung
City, Taiwan, 26-28 Sept. 2012, pp. 252-257.
[77] Sorin, S.: On repeated games with complete information. Mathematics of Operations Research 11(1), 147-160 (1986).
[78] Mailath, G.J., Obara, I., Sekiguchi, T.: The maximum efficient equilibrium payoff in the repeated prisoners dilemma. Games and Economic
Behavior 40(1), 99-122 (2002).
[79] Mailath, G.J., Matthews, S. A., Sekiguchi, T.: Private strategies in finitely repeated games with imperfect public monitoring. Contributions to
Theoretical Economics 2(1), 1-2 (2002).
[80] Kandori, M.: Randomization, communication, and efficiency in repeated games with imperfect public monitoring. Econometrica 71(1),
345-353 (2003).
[81] Green, E.J., Porter, R. H.: Noncooperative collusion under imperfect price competition. Econometrica 52(1), 87-100 (1984).
[82] Srivastava, V., Dasilva, L. A.: Equilibria for node participation in ad hoc networks - an imperfect monitoring approach. In: Proceedings of
IEEE International Conference on Communications (ICC), Istanbul, Turkey, 11-15 June 2006, pp. 3850-3855.
[83] Bhaskar, V., Eric, V. D. : Moral hazard and private monitoring. Journal of Economic Theory 102(1), 16-39 (2002).
[84] Compte, O.: On failing to cooperate when monitoring is private. Journal of Economic Theory 102(1), 151-188 (2002).
[85] Kandori, M.: Introduction to repeated games with private monitoring. Journal of Economic Theory 102(1), 1-15 (2002).
[86] Aoyagi, M.: Collusion in dynamic bertrand oligopoly with correlated private signals and communication. Journal of Economic Theory 102(1),
229248 (2002).
[87] Compte, O.: Communication in repeated games with imperfect private monitoring. Econometrica 66(3), 597626 (1998).
[88] Kandori, M., Matsushima, H.: Private observation, communication and collusion. Econometrica 66(3), 627652 (1998).
[89] Rebahi, Y., Mujica-V, V. E., Sisalem, D.: A reputation-based trust mechanism for ad hoc networks. In: Proceedings of the 10th IEEE

26

Symposium on Computers and Communications (ISCC), Cartagena, Spain, 27-30 June, 2005, pp 37-42.
[90] Samian, N., Seah, W. K. G., Chen, G.: Quantifying selfishness and fairness in wireless multihop networks. In: Proceedings of 38th Annual
IEEE Conference on Local Computer Networks (LCN), Sydney, Australia, 21-24 Oct. 2013, pp. 270-278.

Normalia Samian is a tutor at the Faculty of Computer Science and Information Technology,
Universiti Putra Malaysia (UPM). She obtained her Diploma (Engineering), B.Sc (Computer
Science) from UPM where she received Best UPM Student Award in 2006. In 2010, she
received her M.Sc (Computer Science) from Universiti Teknologi Malaysia (UTM) in the
area of Wireless Networks Security and was nominated for Best Academic Award candidate.
Currently, shes pursuing her Ph.D degree in UPM in the area of cooperation in wireless
multihop networks. Her research interests include ad hoc networks security, trust management
in MANETs and game theoretical approaches for wireless multihop networks.
Zuriati Ahmad Zukarnain is an Associate Professor at the Faculty of Computer Science and
Information Technology, University Putra Malaysia (UPM) Malaysia. She is the head for
High Performance Computing Section at Institute for Mathematics and Research (INSPEM),
University Putra Malaysia. She received her PhD from the University of Bradford, UK. Her
research interests include: Efficient multiparty QKD protocol for classical network and cloud,
load balancing in the wireless ad hoc network, quantum processor unit for quantum computer,
Authentication Time of IEEE 802.15.4 with Multiple-key Protocol, Intra-domain Mobility
Handling Scheme for Wireless Networks, Efficiency and Fairness for new AIMD Algorithms
and a Kernel model to improve the computation speedup and workload performance. She has
been actively involved as a member of the editorial board for some international peer-reviewed
and cited journals. Dr. Zuriati is currently undertaking some national funded projects on QKD
protocol for cloud environment as well as routing and load balancing in the wireless ad hoc
networks.
Winston K. G. Seah received the Dr.Eng. degree from Kyoto University, Kyoto, Japan, in
1997. He is currently Professor of Network Engineering in the School of Engineering and
Computer Science, Victoria University of Wellington, New Zealand. Prior to this, he has
worked for more than 16 years in mission-oriented research, taking ideas from theory to
prototypes, most recently, as a Senior Scientist (Networking Protocols) in the Institute for
Infocomm Research (I2R), Singapore. He is actively involved in research in the areas of
mobile ad hoc and sensor networks, and co-developed one of the first Quality of Service
(QoS) models for mobile ad hoc networks. His latest research interests include wireless sensor
networks powered by ambient energy harvesting (WSN-HEAP), wireless multi-hop networks,
software defined networking, and mobility-enhanced protocols and algorithms for networked
swarm robotics and sensing applications in terrestrial and oceanographic networks. He is a
Senior Member of the IEEE. Homepage: http://www.ecs.vuw.ac.nz/ winston/.

Azizol Abdullah obtained his Master of Science in Engineering (Telematics) from the University of Sheffield, UK in 1996 and his PhD in Distributed System from Universiti Putra
Malaysia, Malaysia in 2010. He is a Senior Lecturer at Department Communication Technology and Network, Faculty of Computer Science and Information Technology, Universiti
Putra Malaysia and currently hold the position of Deputy Dean of Academics and Student Affairs. His main research areas include cloud and grid computing, network security, wireless
and mobile computing and computer networks.

27

Zurina Mohd Hanapi received her first degree in BSc. Computer and Electronic System from
the University of Strathclyde, Glasgow, UK in 1999. The author then received her master
degree in MSc. Computer and Communication System Engineering from Universiti Putra
Malaysia (UPM), Selangor, Malaysia in 2004 and Ph.D from the Universiti Kebangsaan
Malaysia (UKM) in 2011. Currently she is a senior lecturer in the Department of Communication Technology and Network, Faculty of Computer Science and Information Technology,
UPM. She had published more than 20 papers in the cited journals and conferences in the area
of security and wireless sensor network. Her current research interests are on security, routing,
wireless sensor network, wireless network, distributed computing, and cyber-physical-system.
Dr. Zurina is a member on Malaysian Security Committee Research (MSCR) and IEEE.

28